Data mining of biological data for pathway-level analysis
The system addresses inefficiencies in conventional biological data analysis by generating parallel fold-change plots and integrating AI for validated pathway-level insights, improving the efficiency and accuracy of multi-condition data interpretation.
Patent Information
- Application Number
- PCT/US2026/027193
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-07
- Filing Date
- 2026-05-07
- Publication Date
- 2026-11-12
AI Technical Summary
Conventional methods for analyzing biological data across multiple conditions are impractical and error-prone due to the exponential increase in pairwise comparisons, lack of integrated pathway visualization, and reliance on manual inference of cross-condition trends, which limits the efficiency and accuracy of data interpretation.
A system and method for generating parallel fold-change plots and pathway-level analytical outputs using artificial intelligence, integrating fold-change values with standardized biological databases, and enabling interactive visualization and validation to facilitate efficient comparison and interpretation of biological data across multiple conditions.
Enables simultaneous comparison and visualization of biological data across multiple conditions, providing structured pathway-level insights with reduced manual effort and improved accuracy through AI-assisted validation, enhancing the efficiency and reproducibility of biological data analysis.
Smart Images

Figure US2026027193_12112026_PF_FP_ABST
Abstract
Description
Patent ApplicationAttorney Docket 136766.8009. WO00DATA MINING OF BIOLOGICAL DATA FOR PATHWAY-LEVEL ANALYSISCROSS-REFERENCE TO RELATED APPLICATION
[0001] This application claims the benefit of priority of U. S. Provisional Patent Application No.63 / 801,266, filed on May 7, 2025. The entire contents of this Provisional Patent Application is incorporated by reference in its entirety herein.TECHNICAL FIELD
[0002] This patent document relates to a computer platform that is configured to perform data mining of biological measurement data, and more particularly to systems and methods for multi-condition comparison, and pathway-level interpretation of quantitative omics data.BACKGROUND
[0003] Biological research generates quantitative measurement data from experiments involving multiple sample groups, conditions, or treatments. Various statistical techniques and visualization methods have been developed to compare such data across sample groups. One approach involves calculating foldchange values, which measure the magnitude of differences in biological entity levels between sample groups. Tire significance of these differences is evaluated using measures such as p-values. A widely used visualization method is the volcano plot, which represents fold-change and significance as a scatter plot, allowing identification of biological entities that exhibit significant differential expression between two conditions.
[0004] However, these conventional approaches are limited to comparing two conditions at a time. When experiments involve multiple conditions, each pairwise comparison must be inspected separately, and cross-condition trends must be inferred manually. As the number of conditions increases, the number of individual comparisons grows substantially, making manual analysis impractical and error-prone. Furthermore, conventional visualization methods present biological entities independently without alignment to biological pathway relationships, and existing pathway visualizations are generally static and descriptive. There is a need for improved systems and methods that enable efficient comparison of biological data across multiple conditions, visualization of cross-condition patterns within biological pathway contexts, and generation of pathway-level analytical outputs from such data.SUMMARY
[0005] Disclosed are devices, systems, and methods for analysis, visualization, and pathway-level interpretation of biological measurement data across multiple sample groups.187285220.1
[0006] In one example aspect, a method of providing interactive data analysis is disclosed, performed by a computer comprising one or more processors, comprising: receiving an input comprising a plurality of sample groups comprising experimental data, wherein each sample group corresponds to data obtained in an experiment using a particular value of a feature from a plurality of features; calculating, for the input, fold-change values representing changes among the sample groups; determining a number of labels representative of the fold-change values, wherein each label identifies two sample groups; mapping the plurality of features along a first direction of a graph; mapping the labels along a second direction of the graph; depicting, for each feature, a corresponding curve, wherein data points of the curve correspond to fold-change values for the feature across the labels; associating each data point with a corresponding pathway map that depicts a pathway; receiving a user selection of one or more specific data points; obtaining, by retrieving from a database, graphical information of a pathway corresponding to the one or more specific data points; and making an interactive visual presentation of the pathway on a display interface such that the visual presentation identifies features in the pathway and their deviations during the experiment.
[0007] In another aspect, a method of providing interactive data analysis is disclosed, performed by a computer comprising one or more processors, comprising: receiving an input comprising a plurality of sample groups comprising experimental data, wherein each sample group corresponds to data obtained in an experiment using a particular value of a feature from a plurality of features; calculating fold-change values representing changes among the sample groups; determining a number of labels representative of the fold-change values, wherein each label identifies two sample groups; mapping the plurality’ of features along a first direction of a graph; mapping the labels along a second direction of the graph; depicting, for each feature, a corresponding curve, wherein data points of the curve correspond to fold-change values for the feature across the labels; associating each data point with a corresponding pathway map that depicts a pathway; verifying, against a standardized biological database, feature identifiers corresponding to the plurality of features; transforming quantitative data represented in the graph into a structured, machine-readable pathway data object integrating the verified feature identifiers and the fold-change values; supplying the structured pathway data object to an artificial intelligence inference engine to generate a pathway-level analytical output; receiving a user selection of one or more specific data points; obtaining, by retrieving from a database, graphical information of a pathway corresponding to the one or more specific data points; validating the pathway-level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the pathway-level analytical output entities or relationships not present in the structured pathway data object; and making an interactive visual presentation of the pathway on a display interface such that the visual presentation identifies features in the pathway, their deviations during the experiment, and tire pathway-level analytical output.- 2 - 187285220.1
[0008] In another aspect, another data analysis method is disclosed. The method includes computing parallel fold-change plots from sets of experimental data, wherein each of the fold-change plots indicates deviations of one or more features in a set of experimental data; arranging the parallel fold-change plots along a horizontal axis in units of sample change group pairs; and drawing lines connecting data points in the parallel fold-change plots representing a same feature across different sample change group pairs in which the feature is used.
[0009] In another aspect, a system for generating pathway-level analytical outputs from biological measurement data is disclosed. The system comprises one or more processors and a memory' storing instructions that, when executed by the one or more processors, cause the system to: receive quantitative measurement data for a plurality of biological entities across a plurality of experimental conditions; compute fold-change values for each biological entity across the plurality of experimental conditions; identify one or more biological pathways associated with the biological entities; verify, against a standardized biological database, biological entity identifiers corresponding to the biological entities; transform the fold-change values into a structured, machine-readable pathway data object integrating the verified biological entity identifiers and the fold-change values; supply the structured pathway data object to an artificial intelligence inference engine to generate a pathway-level analytical output; and validate the pathway-level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the pathway-level analytical output entities or relationships not present in the structured pathway data object.
[0010] These, and other, aspects are disclosed throughout the present document.BRIEF DESCRIPTION OF THE DRAWINGS
[0011] FIG. 1 depicts an example set up of a biological experiment.
[0012] FIG. 2 depicts an example of a log-ratio plot for comparing biological entities between sample groups.
[0013] FIG. 3 is an example of a volcano plot for visualizing fold-change and statistical significance between conditions.
[0014] FIG.4 is an example highlighting complexity of data analysis due to increased number of pairwise comparisons in comparative analysis.
[0015] FIG. 5 shows another example highlighting complexity of data analysis using volcano plot.
[0016] FIG. 6 shows an example of a rotated volcano plot.
[0017] FIG. 7 shows an embodiment implementing the general structure of FIG. 16, showing an example of a parallel fold-change plot displaying fold-change values for metabolites across multiple experimental comparison conditions.- 3 - 187285220.1
[0018] FIG. 8 (Table 1) shows an example of a data table comprising statistical values associated with a parallel fold-change plot.
[0019] FIG. 9 (Table 2) shows an example of an input data fde format for biological measurement data.
[0020] FIG. 10 shows an example of a selection user interface for generating a parallel fold-change plot.
[0021] FIG. 11 (Table 3) shows an embodiment of an interactive user interface for selecting features using a category-based approach.
[0022] FIG. 12 is an example of an interactive parallel fold-change plot generated by the biological data visualization system based on metabolite data.
[0023] FIG. 13 shows an example of a data table display for detailed statistical values of a particular feature.
[0024] FIG. 14 shows an example of a pathway map displayed by the biological data visualization system.
[0025] FIG. 15 is another example of a pathway map displayed by the biological data visualization system using a different comparison condition.
[0026] FIG. 16 illustrates a general example of parallel comparison of multiple samples.
[0027] FIG. 17 is a workflow for automatically generating tire parallel fold-change plot.
[0028] FIG. 18 depicts a workflow example.
[0029] FIG. 19 depicts a workflow example.
[0030] FIG. 20 illustrates a method for generating a parallel fold-change plot.
[0031] FIG. 21 is a block diagram of an example hardware platform for implementing tire disclosed techniques.
[0032] FIG. 22 is an interactive input user interface for biological data analysis using a metabolomics workflow.
[0033] FIG. 23 is an interactive user interface for generating parallel fold-change plot and displaying pathway information, including enriched pathways and metabolite information associated with metabolic pathways.
[0034] FIG. 24 depicts an interactive user interface designed for selecting sample group comparisons.
[0035] FIG. 25 shows an interactive input user interface for transcriptomics analysis.
[0036] FIG. 26 shows an interactive selection user interface for generating parallel fold-change plot with transcriptomics data.
[0037] FIG. 27 illustrates a system for pathway-centric interpretation of biological measurement data.
[0038] FIG. 28 illustrates a generated parallel fold-change plot with transcriptomics data.
[0039] FIG. 29 illustrates a user interface and interaction flow for generating pathway-centric-4 - 187285220.1interpretations.
[0040] FIG. 30 illustrates a pathway-centric context builder.
[0041] FIG. 31 illustrates a biological database validation module.
[0042] FIG. 32 illustrates a cross-validation module.
[0043] FIG. 33 illustrates a multi-omics integration and correlation analysis module.
[0044] FIG. 34 illustrates an interpretation output module.
[0045] FIG. 35 illustrates a parallel fold-change plot generation engine.
[0046] FIG. 36 illustrates a metabolite-specific structural components of a parallel fold-change visualization for Al-based interpretation.
[0047] FIG. 37 is a block diagram showing some of the components typically incorporated in at least some of the computer systems and other devices on which tire biological data visualization system operates, in accordance with some implementations of the present technology.
[0048] FIG. 38 is a system diagram illustrating an example of a computing environment in which the biological data visualization system operates, in accordance with some implementations of the present technology.
[0049] FIG. 39 illustrates an example configuration window for an artificial intelligence (Al) system.
[0050] FIG.40 illustrates an example interface for receiving experimental metadata associated with input data.
[0051] FIG. 41 illustrates an example interface for selecting a target biological entity category' associated with input data.
[0052] FIG. 42 illustrates an example user interface for receiving a text-based input for interpreting a selected data visualization using a large language model (LLM).
[0053] FIG. 43 illustrates an example interface for selecting a region from a pathway image based on a parallel fold-change plot and performing Vision LLM-based interpretation.
[0054] FIG. 44 illustrates an example window displaying Al-generated results, a follow-up prompt interface for refining initial analysis results, and a corresponding result display screen.
[0055] FIG. 45 is a system architecture diagram illustrating a client device configured to execute an application and a server device including a GPU-based system configured to perform large language model (LLM) inference, wherein the client device communicates with the server device via an application programming interface (API) over a network.Terminology
[0056] In some embodiments, the term "parallel fold-change plot" refers to a visualization comprising: (a) a biological entity axis arranged along a first direction and limited to features associated with a selected - 5 - 187285220.1biological pathway; (b) a plurality of parallel condition axes extending across the graph, each corresponding to a distinct sample group comparison identified by one of the labels; (c) for each feature, a fold-change trajectory connecting data points representing log2 fold-change values across the plurality of condition axes; and (d) visual differentiation attributes applied to distinguish cluster membership or regulatory grouping among the features. See. e.g., FIG. 7, 12, and 16.
[0057] In some embodiments, the term "structured pathway data object" refers to a machine -readable data structure comprising: (i) a pathway identifier for a biological pathway associated with the plurality of features; (ii) validated biological entity identifiers mapped to standardized database entries in at least one of KEGG, HMDB, UniProt, or GO databases; (iii) quantitative statistical metrics comprising log2 foldchange values and fold-change values for each biological entity across the plurality of sample group comparisons; and (iv) experimental metadata comprising genotype information, treatment conditions, and comparison context. See, e.g., FIG. 30.
[0058] In some embodiments, the term "pathway-level analytical output" refers to a natural language interpretation generated by an artificial intelligence inference engine based on the structured pathway data object, wherein the output comprises at least one of: (i) a pathway activation state describing upregulation of biological entities within the selected biological pathway; (ii) a pathway inhibition state describing downregulation of biological entities within the selected biological pathway; or (iii) a regulatory network state inferred from correlation patterns among biological entities and fold-change trajectories across experimental conditions.
[0059] In some embodiments, the term "constraint" refers to at least one validation rule applied to restrict an artificial intelligence inference engine from generating factually incorrect outputs, comprising: (a) exclusion of external knowledge by restricting generation of compounds, reactions, or pathways not present in the structured pathway data object; (b) evidence-based rewriting by correcting or removing statements that contradict the structured pathway data object; or (c) hypothesis marking by labeling statements that are logically inferable but not explicitly supported by the structured pathway data object.
[0060] In some embodiments, the tenn " KEGG REST API" refers to a web service application programming interface provided by the Kyoto Encyclopedia of Genes and Genomes (KEGG) database at https: / / www.kegg.jp / kegg / rest / keggapi.html, enabling programmatic access to KEGG pathway information, biological entity identifiers, reaction connectivity data, enzyme-substrate relationships, and pathway structural data. Tire KEGG REST API supports retrieval of pathway maps, compound information, enzyme classifications, and ortholog annotations for integration with biological measurement data analysis workflows.- 6 - 187285220.1DETAILED DESCRIPTIONInitial Discussion
[0061] With the increasing scale of research and the need to compare multiple sample groups, there is a growing demand for efficient methods to analyze and interpret complex biological comparisons. This document proposes a faster and easier approach for comparing changes in metabolites or transcripts within multiple sample groups, along with an automated algorithm to facilitate the process. The invention also introduces a visualization method for comparing metabolic networks, addressing the challenges posed by the larger scale of research and providing efficient tools for analyzing and interpreting complex biological comparisons.
[0062] The present document introduces a novel plot format referred to as a parallel fold-change plot, and a system or tool for automated generation of the parallel fold-change plot in association with biological pathways, referred to as ShowMEPATH (Show MEtabolic PATHways). Tire disclosed system enables simultaneous comparison of multiple biological samples and visualization of complex biological data.
[0063] The ShowMEPATH automates the process of generating biological pathways, which involves associating biological entities (such as genes, proteins, or metabolites) with specific biochemical or signaling pathways. This automated system reduces the time and effort required for pathway mapping, allowing researchers to gain insights into the functional relationships and interactions among molecules within a pathway context.
[0064] Using ShowMEPATH enables researchers to efficiently explore and interpret complex biological data involving multiple samples, and to ultimately gain valuable insights for further analysis and experimentation.
[0065] In addition, the disclosed system incorporates an artificial intelligence model or a large language model (LLM) to provide interpretative analysis of generated biological data and to support interactive follow-up analysis based on user queries.Technology Introduction
[0066] The metabolome of a biological sample refers to the collection of intermediate and end products present in the metabolic network. Comparing mctabolomcs across different sample groups can provide insights into the underlying structure of the metabolic network through distinct metabolic outputs. To perform such comparisons, various statistical techniques and visualization methods are employed.
[0067] One commonly used approach involves calculating the fold-change, which measures the magnitude of differences in metabolite or transcript levels between sample groups with multiple replicates. Tire significance of these differences is evaluated using measures such as p-values. In the context of metabolomics, a popular visualization method is the Volcano Plot, which represents the fold-change and- 7 - 187285220.1significance of metabolite differences as a scatter plot. This plot allows for the identification of metabolites that exhibit significant differential expression by visually highlighting those with large fold-changes and low p-values. Another visualization approach utilized in ornics data analysis, including metabolomics, proteomics, and transcriptomics, is the Ratio Plot. This plot exclusively displays the log2 fold-change values. It provides a simplified representation of the differences between conditions or groups, facilitating the identification of features that show substantial changes.
[0068] The Volcano Plot has been widely employed for over two decades in the field of omics data analysis. It serves as a usefill tool for statistical analysis and visualization, enabling researchers to assess the significance and extent of changes in gene expression, protein abundance, metabolite levels, or other biological measurements between different conditions or sample groups. By examining the Volcano Plot, researchers can identify, prioritize, and filter features that exhibit significant differential expression for further investigation.
[0069] The Volcano Plot is a scatter plot commonly used in omics data analysis to visualize and compare changes in gene expression, protein abundance, metabolite levels, or other biological features between different conditions. In a Volcano Plot, the x-axis represents the log2 fold-change values, indicating the magnitude of expression change. Features with high positive log2 fold-change values are positioned on the right side, indicating upregulation, while those with high negative log2 fold-change values are located on the left side, representing downregulation. The y-axis of the Volcano Plot represents the statistical significance, often displayed as the -logic, of the p-value or adjusted p-value. Points higher on the plot (greater y-valucs) indicate higher statistical significance, suggesting that the observed differences arc less likely to have occurred by chance.
[0070] The main advantage of the Volcano Plot is its ability to simultaneously display both the magnitude of change and the statistical significance, allowing researchers to identify features that exhibit substantial changes and are statistically significant. By setting appropriate thresholds for fold-change and statistical significance, it becomes easier to identify genes, proteins / peptides, metabolites, or features of interest that are differentially expressed. However, the Volcano Plot is typically limited to comparing two conditions at a time. When experiments involve multiple conditions, each pairwise comparison must be inspected separately, and coordinated cross-condition trends must be inferred manually. As the number of conditions increases, the number of individual comparisons grow s substantially, making manual analysis impractical and error-prone.
[0071] The Volcano Plot is also used for pathway mapping, which involves associating biological entities (such as genes, proteins, or metabolites) with specific biochemical or signaling pathways. This process helps researchers understand the functional relationships and interactions within a pathway and gain insights into the overall biological significance of their experimental findings. However, conventional- 8 - 187285220.1pathway mapping approaches generally produce static visualizations that present biological entities independently without structured alignment to biological pathway relationships across multiple experimental conditions. Although biological entities may be overlaid onto curated pathway diagrams, such visualizations do not generate structured, machine -readable representations encoding multi-condition quantitative vectors, regulatory directionality, or pathway membership constraints within an integrated analytical framework.
[0072] Furthermore, existing systems do not transform pathway-constrained visualization data into structured data objects suitable for direct input into artificial intelligence inference models for generating pathway-level analytical outputs. Nor do conventional platforms integrate multi -omics measurements and experimental metadata into a unified, validated, pathway-restricted interpretive context. While Al-assisted interpretation provides automated analysis capabilities, reliance solely on Al inference may introduce factual inaccuracies (hallucination), where plausible but incorrect statements are generated regarding compounds, reactions, or pathway relationships not supported by the underlying experimental data or authoritative biological databases. Conventional systems present Al -generated interpretations without mechanisms for researcher verification, leaving users unable to validate whether conclusions align with original quantitative data. To address these limitations, a hybrid approach is introduced that combines automated analysis with human expert verification. This hybrid method presents both pathway-constrained visualizations and corresponding Al-generated interpretations in a side-by-side comparison interface, enabling researchers to verify factual accuracy against original quantitative data, validate statistical significance directly from visual representations, check logical consistency of pathway relationships, and modify or annotate Al-generated statements as needed before final approval. There is therefore a need for systems and methods that enable simultaneous multi-condition comparison, structured pathway-centric data transformation, validated pathway-level interpretation with researcher verification, and hybrid workflows combining automated analysis with human expert review of biological measurement data.AI-Assisted Biological Data Interpretation
[0073] Recent advances in artificial intelligence, particularly transformer-based models such as large language models (LLMs) and vision-language models (VLMs), have demonstrated capability in processing structured textual and visual inputs to generate natural language outputs. These models have been applied in various scientific domains, including biological data analysis, where they may be used to generate interpretive summaries from quantitative experimental data. However, when applied to biological interpretation tasks, such models may generate plausible but factually incorrect statements, a phenomenon commonly referred to as hallucination. In tire context of biological pathway analysis, hallucination may manifest as the generation of compounds, reactions, or pathway relationships that are not supported by the underlying experimental data or authoritative biological databases. Constraining Al-generated- 9 - 187285220.1interpretations to verified biological entities and relationships remains a significant technical challenge.
[0074] Conventional approaches to biological data analysis typically present quantitative results in tabular or graphical formats intended for human inspection. These formats arc not structured for direct processing by Al inference models, which may benefit from organized, machine-readable input that encodes pathway identifiers, validated biological entity identifiers, quantitative metrics, and experimental metadata in a unified representation. The absence of structured, pathway-centric data objects in existing workflows creates a gap between quantitative biological measurement data and Al-assisted mechanistic reasoning. Transformation pipelines that convert pathway-constrained visualization data into structured data objects suitable for input to Al inference engines may address this gap.
[0075] Vision-language models (VLMs) are Al models capable of processing both image data and textual input to generate contextual outputs. In biological research, pathway diagrams such as those available from the KEGG database represent complex networks of biochemical reactions, enzyme relationships, and metabolite interactions in a visual format. Conventional systems require manual identification and selection of entities within such diagrams. VLMs may be applied to process user-selected regions of pathway images to automatically identify and standardize biological entities depicted within tire selected region, enabling image-based construction of interpretation contexts. Systems that support such multimodal interaction, combining both text-based and image-based input paths, may enable more intuitive construction of pathway-centric analytical contexts.
[0076] Reproducibility of ALgenerated biological interpretations presents a further technical challenge, as stochastic variation inherent in generative Al models may produce different outputs from identical inputs across successive invocations. Existing systems do not address this variability in the context of biological pathway interpretation. Cross-validation architectures that execute multiple independent inference processes and compare outputs may improve interpretive consistency. Such architectures may apply constraint-based fact checking against authoritative biological database records, including exclusion of externally sourced knowledge, evidence-based correction of contradictory’ statements, and explicit marking of hypothetical inferences, to improve robustness and reproducibility of pathway-level analytical outputs.
[0077] Al inference models, including transformer-based LLMs, operate within finite context windows that limit the volume of input data that can be processed in a single invocation. When biological pathway analysis involves varying numbers of compounds, the volume of associated database information, including reaction connectivity, enzyme -substrate relationships, and pathway structural data, may vary' substantially. Supplying excessive context may degrade inference quality, while insufficient context may limit the depth of interpretation. Dynamic adjustment of context size based on input data characteristics, such as scaling the volume of retrieved biological database information in proportion to the number of compounds under analysis, may balance the depth of biological context with the processing capacity of the inference engine.- 10 - 187285220.1
[0078] While Al-assisted interpretation provides automated analysis capabilities, reliance solely on Al inference may introduce factual inaccuracies (e.g., hallucination). To address this limitation, a hybrid method is introduced that combines automated Al analysis with human expert verification through a side-by-side comparison interface, enabling researchers to verify accuracy against original quantitative data, validate statistical significance directly from plots, and approve final interpretations before publication or further analysis.Example Comparing Phenotypes in Biological Experiments
[0079] FIG. 1 depicts an example setup of a biological experiment. FIG. 1 presents representative images illustrating the varied responses of wild type (WT), mutant (cog 7), and the mutant line complemented with the native COG 7 genes (CP) to dark treatment (1.1). Dark treatment induces sugar deprivation, a process partially reversible by the addition of sugar (sucrose) to the media.
[0080] The images depict accelerated senescence in the cog7 mutant compared to other genotypes in the absence of added sucrose to the media (-). However, this phenotype is eliminated in the lower panel in the presence of sucrose (+). Below the images are senescence phenotypes of individual leaves detached from the three genotypes, highlighting accelerated senescence in cog7 leaves compared to the WT and complemented line (1.3).
[0081] It’s worth noting that COG 7 is a component of the Golgi apparatus, playing a central role in controlling Golgi-related membrane protein trafficking and localization (1.2). These experiments will generate data suitable for analysis through metabolomics, proteomics, or transcriptomics techniques, facilitating comparative studies.
[0082] The data may need to be expanded by using metabolomics, proteomics, or transcriptomics analyses. Typically, a log-ratio plot or a volcano plot is employed for comparative analysis.
[0083] One technical challenge in these experiments is effectively comparing sample groups. Two frequently employed techniques for this purpose are the log-ratio plot and the volcano plot.Log-Ratio Plot Example
[0084] FIG. 2 depicts an example of a log-ratio plot for comparing biological entities between sample groups. In one use-case. FIG. 2 can illustrate a log-ratio plot example in metabolomics. The x-axis (2.1) represents the mean ratio fold-change (plotted on a log2 scale) of the relative abundance of each metabolite between two selected samples (Sample A (cog7 at 0 days) and Sample B (cog7 at 3 days)). The y-axis (2.2) displays each analyzed metabolite. Metabolites with unchanged abundance between the two samples are positioned at the x-axis origin (indicated by the green-blue vertical line, representing a ratio fold-change of 1). Metabolites that hyper-accumulate in one of the two samples under analysis will plot either to the left or the right of the x-axis origin. The chemical classification of metabolites is presented in plot legends- 11 - 187285220.1(2.3).Volcano Plot Example
[0085] FIG. 3 is an example of a volcano plot for visualizing fold-change and statistical significance between conditions. The x-axis (3.1) displays the mean ratio fold-change (plotted on a log2 scale) of the relative abundance of each metabolite between two selected samples (mirroring the ratio plot in FIG. 2). The y-axis (3.2) represents the statistical significance p-value of the ratio fold-change for each metabolite. Metabolites with unchanged abundance between the two samples are positioned at the x-axis origin (3.4). The blue vertical line signifies a log2 fold-change equal to 1.0. Metabolites that hyper-accumulate in one of the two samples under analysis will be situated to the left or right of the x-axis origin (3.4). The order of metabolites on the y-axis is determined by the statistical significance p-value of the ratio fold-change, typically displayed as the -logic of the p-value on the y-axis. The chemical classification of metabolites is presented in plot legends (3.5).Calculation of Mean, Log fold-change, and Fold-Change (fold-change) in Plots
[0086] Generally, the calculation of the mean values for sample groups, denoted as fisland ps2. using the following formulas: There are two methods to calculate the mean value used in log2fold-change values, as follows.in which pslrepresents the mean of the first samples group (Si). It is calculated by taking the sum of all individual values (Su) in the first sample group (Si) and dividing it by the total number of values (n) in that group. Similarly, fis2represents the mean of the second sample group (S2). It is calculated in the same way as / Tsl, but using the values from the second sample group (S2i). Thus, / rsland fis2are the mean values of the first and second sample groups (Si and S2), respectively, and are calculated by averaging the values within each group.
[0087] Then, the log2 fold-change (log2fold-change) is calculated between two sample groups fisland qs2) as follows.Log2fold-change = log2(si / s2) (3) Log2fold-change = log2( Ts2) - log2( rsi) (4) in which the log2fold-change is calculated by taking the ratio of the mean expression values in the first sample group (Si) to the mean expression values in the second samples group (S2). It represents the log2fold-change between the two groups. This alternative way for log2fold-change calculates the difference between the log2-transformed mean expression values of the second sample group (S2) and the first sample- 12 - 187285220.1group (Si). It provides the same information as the first formula but expressed in terms of the difference in log2-transformed values. In both formulas,S1ands2are the mean values of the first and second sample groups (Si and S2), respectively. The log2 transformation is commonly used in such analyses to make foldchanges more interpretable and to linearize the data.
[0088] To calculate the mean value used in log2fold-change values, the alternative method suggests using the mean of log-transformed values due to the characteristics of the lognormal distribution. As the standard deviation of the lognormal distribution increases, the error in calculations based on raw values grows. Therefore, taking the mean after log transformation is considered consistently superior and it is beneficial for analyzing data with low sample reproducibility.in which the mean (μsl) of the log2-transformed values in the first sample group (Si) is calculated.
[0089] Similarly, the mean ( / 7s2) of the log2-transformed values in the second sample group (S2) is calculated.Log2fold-change = ( / ?sl) / ( / ?s2) (7)
[0090] Then, finally, the log2fold-change by taking the ratio of the mean log2-transformed values in the first sample group (Si) to those in the second sample group (S2) calculates.Log2fold-change
[0091] It is alternative formula for Log2fold-change calculates the difference between the mean log2-transformed values of the second sample group (S2) and the first sample group (Si). Thus, the alternative method involves log -transforming the data before calculating the mean, and Log2fold-change represents the ratio or difference of these mean log-transformed values between two sample groups. This approach is used to handle data with a lognonnal distribution more effectively.Fold-change (fold-change) = 21’A"1 d’change value(9) in which the fold-change (fold-change) is calculated by raising 2 to the power of the Log2fold-change value. It is based on the understanding that the fold-change is inherently related to exponential growth or decay.Examples of Technical Problems
[0092] FIG. 4 serves as an illustrative example of a comparison involving multiple groups, generating a substantial number of individual comparisons (276 in this particular scenario), each visualized as a two-dimensional plot. The complexity arises from the fact that manually examining and discerning specific changes in metabolites across all these comparisons is both impractical and susceptible to errors.
[0093] To delve deeper into the challenge, consider a dataset that encapsulates variations in the metabolites of a specific enzyme employed in the experiment. Attempting to navigate through this extensive array of comparisons manually would be a daunting task, given the intricate nature of the data and the- 13 - 187285220.1potential for oversight or inaccuracies. The sheer volume of comparisons necessitates more sophisticated analytical approaches, such as automated algorithms or computational methods, to derive meaningful insights from the intricate patterns and variations within the dataset. In some embodiments, artificial intelligence-based interpretation or large language model (LLM)-assisted analysis may be utilized to facilitate interpretation of complex biological relationships and support follow-up analytical queries associated with generated analysis results.
[0094] FIG. 5 highlights the challenge of comparing large numbers of chemical compounds through volcano plots with real complex mixture examples. Such complexity is common in fields like transcriptomics, proteomics, untargeted metabolomics, petroleomics, and other omics studies. Volcano plots are sometimes utilized to filter chemical compounds and reveal quantitative changes in major classes within complex mixture samples such as crude oil, natural organic matter (NOM), and dissolved organic matter (DOM) obtained through ultra-high resolution mass spectrometry techniques.
[0095] As the ratio proportion of sample A increases, both positive and negative fold changes exhibit constant variation, demonstrating a clear trend in major classes. However, the identification of changes in individual compounds within numerous volcano plots poses significant challenges.
[0096] FIG. 5 exemplifies a vertically stacked volcanic plot, providing a visual representation of the intricate comparisons involved in studying complex mixtures, such as those found in crude oil, NOM, and DOM samples.
[0097] FIG. 6 shows an example of a rotated volcano plot and illustrates an alternative approach to data presentation where the conventional volcano plot is rotated by 90 degrees. In this configuration, rather than being vertically stacked, the data is represented in sequential steps from left to right (alternatively, a right-to-left presentation may also be utilized). The bold vertical lines serve as demarcations, distinctly separating each volcano plot.Example Comparative Analysis with Fold-Change Analysis by Metabolomics Workflow
[0098] FIG. 7 shows an example of a parallel fold-change plot.
[0099] FIG. 8 (Table 1) shows an example of a data table comprising statistical values associated with a parallel fold-change plot.
[0100] FIG. 9 (Table 2) shows an example of an input data fde format for biological measurement data.
[0101] FIG. 10 illustrates an example user interface, in a state prior to loading input data, provided for illustrating generation of a parallel fold-change plot.
[0102] FIG. 11 (Table 3) shows an embodiment of an interactive user interface for selecting features using a category-based approach.
[0103] FIG. 12 is an example of an interactive parallel fold-change plot based on metabolomics data- 14 - 187285220.1generated by the biological data visualization system.
[0104] FIG. 13 shows an example of a data table display for detailed statistical values of a particular feature.
[0105] FIG. 14 shows an example of a pathway map displayed by the biological data visualization system, wherein fold-change values are represented using colors.
[0106] FIG. 15 is another example of a pathway map displayed by the biological data visualization system using a different comparison condition, wherein fold-change values are mapped to colors for visualization.
[0107] FIG. 7 shows an example of a parallel fold-change plot displaying fold-change values for metabolites across multiple experimental comparison conditions. The plot includes a metabolite axis arranged vertically along the left side of the graph, listing metabolite names. A series of parallel vertical axes extend horizontally across the graph, each corresponding to a distinct experimental comparison condition labeled at the top, with comparison names indicating different genotype and treatment combinations. A horizontal zero baseline separates a positive fold-change region from a negative foldchange region, indicating directionality of regulation. For each metabolite, a fold-change trajectory is generated by connecting quantitative log2 fold-change values across the condition axes, forming multidimensional regulation profiles displayed as colored lines. The trajectories are rendered in different colors to denote cluster membership or regulatory grouping, with a legend indicating that green represents positive log2FC values, gray represents no difference, and orange represents negative log2FC values.
[0108] FIG. 8 shows an example of a data table comprising statistical values associated with a parallel fold-change plot. The table includes a feature name section (Tl-1) displaying the names of selected features. Sample labels (T1 -2) identify the sample groups used in the parallel comparison. A log2fold-change column (Tl-3) displays the calculated log2fold-change values for each parallel comparison. A fold-change column (Tl-4) displays the corresponding fold-change values. A p-value column (Tl-5) displays the calculated p-values indicating statistical significance for each comparison. An adjusted p-value column (Tl-6) displays the Benjamini-Hochberg adjusted p-values. Table cell values (Tl-7) populate the cells within the columns to display the actual numerical values for each sample comparison. A sample presence column (Tl-8) indicates the presence of samples in the comparison. The table is organized with rows corresponding to different sample label comparisons from Label 1 through Label N.
[0109] FIG. 10 illustrates an example of a selection user interface for generating a parallel fold-change plot. The interface includes a control element for loading new data (3-1) configured to initiate loading of new experimental data for analysis and a control element for loading stored data (3-2) configured to load previously stored experimental analysis data. A display region for enriched pathways (3-3) is configured to present a list of enriched pathways for a plurality of comparison groups. A display region for metabolic - 15 - 187285220.1pathways (3-4) is configured to present metabolic pathways associated with loaded metabolites in a structured form. A display region for metabolites (3-5) is configured to present a list of metabolites included in a selected metabolic pathway. A control element for fold-change plot (3-12) is configured to generate a parallel fold-change plot. A clustering parameter input (3- 13) is configured to receive a clustering parameter for grouping metabolites based on similarity of fold-change patterns using a machine learning model, wherein the clustering parameter is initialized to a default value and is adjustable by a user, and wherein metabolites displayed in region (3-5) are represented in a same color when belonging to a same group. A display region for generated plot (3-14) is configured to present a generated plot. A control element for Al processing (3-15) is configured to initiate processing of an artificial intelligence system based on configured settings. A control element for target selection (3-16) is configured to initiate a selection operation for selecting a target biological entity category associated with input data. A region for displaying interpretation results (3-17) is configured to present interpretation results generated by the LLM. A region for receiving a prompt for follow-up queries (3-18) is configured to receive a prompt associated with follow-up queries with respect to initial analysis results.
[0110] FIG. 16 illustrates the parallel comparison of multiple samples intended to be implemented by some embodiments. This figure consists of several components: (1-1) a section displaying the names of metabolites, transcripts, or features with clickable links to access external databases, showcasing useful information, (1-2) axis lines indicating multiple fold-change values, (1-3) lines representing the log2 scaled fold-change values (log2fold-change), (1-4) linkable labels are provided to access the KEGG pathways for the samples used in the parallel comparison, (1-5) tire positive value section displayed on the fold-change line, and (1-6) the negative value section on the fold-change line. Figure 1 illustrates the patterns of foldchange (log2 scale) variations among Nth parallel comparisons of Feature 2, selected by the user (2-1).
[0111] In particular, log2 fold-change refers to the logarithm (base 2) of the fold-change value between two conditions or groups. The fold-change itself is calculated by dividing the expression level of a particular feature in one condition by its expression level in another condition. Tire logarithm of the fold-change transforms the fold-change values into a logarithmic scale. This transformation allows interpretation of fold-change, statistical analysis and visualization. Additionally, the logarithmic scale can improve tire symmetry of the data distribution and facilitate comparisons between upregulated and downregulated metabolites. A log2 fold-change of 0 indicates no change in expression levels between the two conditions being compared. Positive log2 fold-change values (1-5) signify upregulation. Conversely, negative log2 fold-change values (1-6) indicate downregulation, implying a decrease in expression. The lines (1-3) can be color-coded to distinguish metabolites or features.
[0112] Table 1 (FIG. 8) is a data table that could be activated by the section displaying the selected features’ lines (1-3) in FIG. 16, chosen by the user. This table is used to verify tire actual values for- 16 - 187285220.1biological comparisons involving multiple samples. The table consists of the following components: (Tl-1) the section displaying the names of the selected features by the user, (Tl-2) labels of the samples used in (1-4) of FIG. 16, (Tl-3) log2fold-change calculated for parallel comparison, (Tl-4) fold-change (foldchange), (Tl-5) P-value, (Tl-6) Adj. P-value (B. H), (Tl-8) Sample Presence and finally, (Tl-7) each cell of the table displays the actual values (Tl-3). (Tl-4). and (Tl-5). Here is a detailed description of each component:
[0113] The log2fold-change (log2fold-change) or log2 Ratio is a statistical measure commonly used in data analysis, particularly in genomics, transcriptomics, proteomics, and metabolomics. It quantifies the relative change in expression levels between two conditions or groups.
[0114] Fold-change (fold-change) refers to the ratio of expression levels between two conditions. It indicates how much the expression of a particular metabolite, or feature has increased or decreased between the two conditions. To convert a log2fold-change value to fold-change, the following formula can be used:
[0115] Fold-change (fold-change) = 2 Log fold-change value
[0116] In this formula, the base of the exponentiation is 2, as the log2 fold-change represents the logarithm to tire base 2 of the fold-change value. By raising 2 to the power of the log2 fold-change, we obtain the corresponding fold-change value.
[0117] P-value is typically obtained through statistical analysis, such as hypothesis testing, to assess the significance of differences between two conditions or groups. The specific method for calculating the p-value depends on tire statistical test used. For example, in differential expression analysis, commonly used statistical tests include t-tests, one-, two-, or multi-way ANOVA (Analysis of Variance) with Post hoc testing such as Tukey HSD (Honest Significant Difference), or the Wilcoxon rank-sum test. These tests compare the expression levels or measurements between two groups and calculate a p-value that represents the probability of observing the observed differences or more extreme differences between tire groups by chance.
[0118] Adj. P-value (B. H: the Benjamini-Hochberg procedure) is a statistical method commonly used in multiple hypothesis testing to control the false discovery rate (FDR). Tire BH procedure is widely used in various fields, including genomics, proteomics, and metabolomics, to address the issue of multiple hypothesis testing. It allows researchers to control the FDR while maximizing the detection of truly significant findings.
[0119] Table 2 (FIG. 9) represents the input data used in the implemented system of this invention. The input data should contain at least the following metadata information for each compound: Compound ID (T2-1), Mass (T2-2), Standard Name of the metabolite (T2-3), Formula (T2-4), and KEGG ID (T2-5) for metabolomics. And multiple sample groups (T2-6) for comparison and measurement values from a mass spectrometry (T2-7) for calculating Fold-change and P-value are required. The sample name consists of a - 17 - 187285220.1main group and a sub-group within each group, connected by an underscore
[0120] FIG. 10 illustrates an example of a user interface for implementing multiple-group comparison functionality and multiple-docking windows with AI-features. Tire user interface may illustrate a process in which multiple biological comparison analyses are performed after loading an input table derived from actual experiments. In an embodiment, the prototype may be implemented as a standalone program executable on a Windows operating system and may be developed using the C# language in combination with R packages. In addition, the generated plot may be implemented to operate on a web browser based on HTML and JavaScript, thereby providing cross-platform compatibility across Windows, Linux, and macOS environments. The tool may be configured to generate a parallel fold-change plot for comparing multiple groups and to perform pathway mapping analysis using metabolites obtained from targeted metabolomics experiments. These are the key components of the prototype:
[0121] A pathway analysis component (3-1) for calculating fold-change values and constructing metabolic pathways using the input file and HMDB database; and a visualization component (3-12) for generating parallel fold-change plots with pathway mapping based on the constructed pathways.
[0122] To provide a detailed explanation of the two components of the prototype, the pathway analysis component (3-1) may calculate log2 fold change and p-value using an input file (Table 2) and at least one of the HMDB database or the KEGG database. The visualization component (3-12) may utilize the calculated fold-change values and pathway information generated by the pathway analysis component to generate a parallel fold-change plot based on biological conditions or experimental design, and to perform pathway mapping using metabolites associated with a selected pathway or module generated by the pathway analysis component and a biological comparison selected by a user (3-3-1). Tire generated plots may be displayed in region (3-14). In addition, Al interpretation results may be displayed in region (3-17), and a user prompt for follow-up queries based on the Al interpretation results may be received through region (3-18).
[0123] Table 3 (FIG. 11) is a table representing a user interface schema enabling users to efficiently choose features within a specific Main- and Sub-category, select multiple features and generate a parallel fold-change plot. The final results produced by the preparation phase (2-1) can be presented in two tables, as depicted in Table 3. (T3-1) comprises a group of features such as metabolites or transcripts, and upon selecting a specific group in (T3-1). In Table 3, the Main-category (T3-1-2), Sub-category (T3-1-3), all feature counts in the sub-category (T3-1-4), and the count of found features (T3-1-5) in a particular input experiment can be displayed. Hie Main-category (T3-1-2) and Sub-category (T3-1-3) will vary depending on the omics workflow selected by the user. (T3-2) configures the table using Feature ID (T3-2-2), Name (T3-2-3), other metadata (T3-2-4) such as metabolite properties (elemental formula, InChlKey. SMILES, etc.) for metabolomics workflow and external database IDs (KEGG, UniProt, PF AM, SwissProt, NCBI,- 18 - 187285220.1Gene Ontology (GO), etc.) for proteomics and transcriptomics from the features within the chosen group as dependencies and utilizing this category-based method provides an efficient approach to generate a parallel fold-change plot with numerous features.
[0124] FIG. 17 is a workflow of the biological data visualization system that automatically generates the parallel fold-change plot shown in FIG. 16, which is aimed to be implemented according to embodiments. It can be divided into two main parts: the preparation phase (2-1) and the plot generation phase (2-2).
[0125] As depicted in FIG. 18 and FIG. 19, specifically, the preparation phase (2-1) includes the data input stage (2-1-1), the calculation stage for Foldchange, and P-Value / Adj. P-Value (B. H) (2-1-2), the pathway ID search stage for pathway mapping (2-1-3), the Online KEGG (Kyoto Encyclopedia of Genes and Genomes) database search stage (2-1-4), and the step of combining the pathways obtained from the previous two stages (2-1-3 and 2-1-4) with the metabolites obtained from the data input stage, (2-1-1), (2- 1-5). Finally, the resulting list of identified metabolic pathways / modules with their corresponding metabolites from an experiment is displayed in the table (2-1-6, T2.1 in FIG. 9). The plot generation phase (2-2) consists of the stage where the user inputs the desired information (2-2-1), followed by the process of finding a pathway or a module, which is a component of a pathway. (2-2-2) and concludes with the actual plot generation stage (2-2-3).
[0126] More specifically, in the data input stage (2-1-1), a data table (Table 2) is used. And, during the pathway ID search stage (2-1-3) for Pathway mapping. To find the pathway IDs, the HMDB (the Human Metabolome Database - https: / / hmdb.ca / ) database is utilized. The HMDB database comprises a vast collection of public database reference IDs and other useful metadata, such as KEGG pathways, chemical classification and chemical properties (elemental formula, InChlKey, SMILES, etc.), can be utilized in (3- 2-1, 3-3-3) for thousands of compounds. In this stage, the KEGG IDs associated with the metabolites obtained from the experiment are used to retrieve all the corresponding KEGG pathways within the HMDB database. These KEGG pathways including KEGG IDs provide access to pathway information in the KEGG database, allowing for comprehensive pathway mapping. This approach has the advantage of utilizing alternative names, aliases, synonyms, or other known names of compounds within tire HMDB database to find the correct KEGG pathways when the user does not provide KEGG IDs although there are duplicate IDs present in the pathway list. This is especially advantageous in cases where compounds within the KEGG pathways database can potentially result in incorrect KEGG IDs due to the presence of isomeric compounds, thus allowing for cross-validation. By considering these alternative names, the accuracy of the pathway mapping process is improved. Furthermore, since the KEGG database frequently updates and newly adds pathways information including compounds, the Online KEGG database search stage (2-1-4) is performed to access the latest information from the KEGG database using KEGG API, as defined herein, and validating the pathways including compounds from HMDB database for the- 19 - 187285220.1MetabolomicsTargetedWorkflow (3-1-1). The TranscriptomicsWorkflow uses the KEGG API, as defined herein, but pre-downloaded files for enzymes in the KEGG database can also be used to reduce runtime.
[0127] FIG. 20 illustrates a method (2000) for generating a parallel fold-change plot. The method comprises: computing (2002) parallel fold-change plots from sets of experimental data, wherein each of the fold-change plots indicates deviations of one or more features in a set of experimental data; arranging (2004) the parallel fold-change plots along a horizontal axis in units of sample change group pairs; and drawing (2006) lines connecting data points in the parallel fold-change plots representing a same feature across different sample change group pairs in which the feature is used. In some embodiments, the features correspond to metabolites obtained from targeted metabolomics experiments. In other embodiments, the features correspond to transcripts obtained from transcriptomics experiments. In further embodiments, the features correspond to proteins or peptides obtained from proteomics experiments. In some embodiments, the fold-change plots are calculated using a logarithmic difference (e.g., Iog2 fold-change) in experimental data collected for sample groups in a sample change group pair. The lines connecting data points may be color-coded to distinguish different features, wherein each feature is assigned a unique color to enable visual identification of fold-change patterns across the sample change group pairs. In some embodiments, the method further comprises associating each data point with a corresponding pathway map that depicts a biological pathway. In response to a user selection of one or more specific data points, graphical information of the pathway corresponding to the selected data points may be retrieved from a database and presented as an interactive visual presentation on a display interface. The interactive visual presentation may identify features in the pathway and their deviations during the experiment. In some embodiments, the method further comprises displaying a data table in response to a user selection of a specific data point, w herein the data table comprises at least one of a log2 fold-change value, a fold-change value, a p-value, or an adjusted p-value for the feature corresponding to the specific data point. Rows in the data table corresponding to statistically significant changes may be visually highlighted. In some embodiments, the method further comprises verifying feature identifiers against a standardized biological database, transforming quantitative data from the parallel fold-change plot into a structured, machine-readable pathway data object integrating the verified feature identifiers and the fold-change values, and supplying the pathway data object to an inference engine to generate a pathway-level analytical output. The pathway -level analytical output may be validated by applying at least one constraint restricting the output to entities and relationships present in the pathway data object.
[0128] FIG. 22 illustrates an example of an interactive input user interface for biological data analysis. The interface may include an OmicsWorkflow (3-1-1) selection region configured to provide workflow options for the analysis. A comparative analysis setup region (3-1-3) may be configured to enable-20 - 187285220.1configuration of sample group comparisons. The interface may further include a main group selection field (3-1-4) and a subgroup selection field (3-1-5), each configured to allow a user to define a grouping structure for comparative analysis. An add control element (3-1-6) may be configured to add configured comparisons to the analysis. A list of sample group comparisons (3-1-7) may be configured to display configured pairwise comparisons created for the analysis. A confirmation control element (3-1-8) may be configured to confirm and execute the configured comparative analysis task.
[0129] OmicsWorkflow (3-1-1) may provide workflow options including a MetabolomicsTargetedWorkflow configured to support targeted metabolomics analysis when KEGG identifiers are available, and a TranscriptomicsWorkflow configured to support transcriptomics analysis.
[0130] OmicsWorkflow (3-1-1) provides workflow options including a MetabolomicsTargetedWorkflow that supports targeted metabolomics analysis when KEGG IDs are available, and a TranscriptomicsWorkflow that allows transcriptomics analysis.
[0131] For configuration of a comparative analysis (3-1-3), sample names included in an input file may be divided into a main group (3-1-4) and a subgroup (3-1-5). A user may configure a comparison between sample groups by combining two separate groups (3-1-4, 3-1-5) for comparative analysis using an add control element (3-1-6). The sample group comparisons (3-1-7) may be configured to display generated pairwise comparisons. After the required parameters are entered, a confirmation control element (3-1-8) may initiate a preparation phase (2-1).
[0132] FIG. 23 illustrates an interactive user interface for generating a parallel fold-change plot and displaying pathway information. The interface includes an enrichment analysis explorer (3-2-1) positioned on the left side, which displays a table containing pathway information with columns for ranking, classification level, pathway identifiers, and pathway names. A pathway explorer section is shown in the center portion of the interface, displaying module information with pathway identifiers and matched counts. A feature explorer (3-2-2) is positioned on the right side of the interface, displaying a table with columns for group index, compound names, and associated metadata. The feature explorer (3-2-2) also includes colored bars representing different metabolite categories.
[0133] FIG. 24 depicts an interactive user interface designed for selecting sample group comparisons. The interface displays a Comparison Analysis Setup window containing a comparison list (3-3-1) that presents multiple sample group comparison pairs available for selection. The comparison list (3-3-1) includes checkboxes adjacent to each comparison entry, allowing users to select specific comparisons for analysis. The comparison entries follow a naming convention indicating control conditions versus experimental conditions across different genotypes. Tire interface includes an Analysis Order section with Move Up and Move Down buttons (3-3-2) that enable users to reorder the selected comparisons. A Select All option is provided to facilitate selection of all available comparisons simultaneously. The interface also- 21 - 187285220.1includes an Apply button and Cancel button for user interaction and confirmation of selections.
[0134] FIG. 12 is a parallel fold-change plot for a particular biological experiment, generated using the prototype software, following the approach presented in FIG. 16. Each metabolite (4-12) is distinguished by its unique color, showcasing the patterns of fold-change for the twenty-five metabolites across twenty-four comparisons using 8 genotypes and 3 treatments (Control (CTRL), Dexamethasone (DEX), and High Light (HL)) in this particular instance (map01070; Biosynthesis of plant hormones). It reveals that there is no difference between two treatments (CTRL and DEX) across all genotypes (4.4.1, 4.5.1, 4.6.1, 4.7.1, 4.8.1, 4.9.1, 4.10.1, and 4.11.1) than other treatments (CTRL vs HL, DEX vs HL). Users can generate for example, FIG. 14 and FIG. 15, which are generated by mouse click events on the labels (4-1 and 4-2) on the generated parallel fold-change plot. These two pathways are considered the same pathway, but by using different fold-change values chosen by the user and coloring them accordingly, allowing users to visually and easily identify the changes in metabolites in their metabolic pathway of interest. Depicted in 4-3, the color-coded representation of log2fold-change on the pathway shown in 6-1 and 6-2.
[0135] FIG. 13 represents the data table that appears when selecting a single biological entity (e.g., ADP-glucose) from FIG. 12. If the P-value and Adj. P-value are less than 0.05 or other significance levels if applicable, each row is highlighted, making it easy to identify sample groups with significant changes. Sample groups that do not show significant changes are labeled “No change” and marked with gray color. Users can define and modify their own colors. Tire table displays columns for Name, log2FC, FC, P, Adj_P (B. H), and Sample Presence. Hie Name column lists comparison conditions between different genotypes and treatments. Rows corresponding to comparisons showing significant changes are highlighted in orange text, while rows showing no significant change are displayed in standard black text.
[0136] In FIG. 13, a data table appears when the user hovers the mouse over a line (1-3) of compounds in an interactive parallel fold-change (fold-change) plot (FIG. 12) generated by the prototype application shown in Figure 3. Tire metabolites (3-5) measured in the experiments, belonging to the selected Pathway / Module (3-4), are color-coded based on their fold-change values for a specific pair of sample groups, as shown in FIG. 12.
[0137] FIG. 14 or FIG. 15, for example, depict the KEGG pathway (map01070; Biosynthesis of plant hormones) on the KEGG website. The URL to access the KEGG module or pathway is generated using ShowMEPATH. These 6-1 and 6-2 are generated by mouse clicks on the parallel fold-change plot described previously, representing color-coded metabolic pathways using different log2fold-change values. Negative log2 fold-change values are displayed in purple, positive log2 fold-change values are represented in green, and areas with no change are depicted in cyan within the selected pathway. In the pathway, a transparent circle or a rectangular box indicates the absence of detected compounds in the experiment.- 22 - 187285220.1Pathway-Centric Transformation and AI-Assisted Interpretation
[0138] In embodiments incorporating artificial intelligence-assisted interpretation, the quantitative data represented in the parallel fold-change plot, including the verified feature identifiers and the fold-change values, are transformed into a structured, machine-readable pathway data object. The pathway data object integrates a pathway identifier for the selected biological pathway, validated biological entity identifiers mapped to standardized database entries, and the fold-change values for each biological entity across the plurality of experimental conditions.
[0139] Prior to transformation, a biological database validation module verifies feature identifiers against a standardized biological database such as KEGG or HMDB. Verifying comprises mapping the feature identifiers to standardized database identifiers and confirming structural association of the features with the selected biological pathway. The validation module retrieves additional relational or structural data from the database, including reaction connectivity and enzyme-substrate relationships, to prepare verified reference information.
[0140] The structured pathway data object can be supplied to an artificial intelligence inference engine to generate a pathway-level analytical output. The pathway-level analytical output may comprise at least one of a pathway activation state (e.g., “Glycolysis pathway is activated under Drug A treatment based on significant positive fold-change values”), a pathway inhibition state (e.g., “TCA cycle is inhibited due to enzyme downregulation reflected in negative fold-change trends”), or a regulatory network state (e.g., “Metabolic reprogramming mechanism inferred from metabolite -enzyme correlation patterns and foldchange trajectories”).
[0141] The pathway-level analytical output can be validated by applying at least one constraint that restricts the artificial intelligence inference engine from including in the pathway-level analytical output entities or relationships do not present in the structured pathway data object. The constraints may comprise at least one of: exclusion of external knowledge by restricting generation of compounds, reactions, or pathways not present in the structured pathway data object: evidence-based rewriting by correcting or removing statements that contradict the structured pathway data object; or hypothesis marking by labeling statements that are logically inferable but not explicitly supported by the structured pathway data object.
[0142] The validated pathway-level analytical output may be presented alongside the interactive visual presentation of the pathway on the display interface, such that the visual presentation identifies features in the pathway, their deviations during the experiment, and the pathway-level analytical output.
[0143] In some embodiments, the method further comprises presenting the pathway-level analytical output alongside the parallel fold-change plot for researcher review and verification, enabling comparison between Al-generated interpretations (3-17) and original quantitative data on the plot (3-14).Example Comparative Analysis by Transcriptomics Workflow- 23 - 187285220.1
[0144] FIG. 25 shows an input interface accessed by clicking the menu (3-1) to load transcriptomics data, including five genotypes and five DPIs (Days Post-Infection): 0, 1, 7, 14, and 22 for each genotype. (7.1) on FIG. 25 indicates that 20 comparative analyses were created using genotypes and DPIs, following a similar approach to the creation of a group comparison list in metabolite analysis (FIG. 22).
[0145] FIG. 26 displays the pathway result (7.2) from the preparation and computation phase (2-1) and highlights the selection of the pathway " Photosynthesis - antenna proteins’ (7.2.1) in Energy metabolism in KEGG pathways. Subsequently, (7.2.2) illustrates transcripts related to the photosystem in the chosen pathway (7.2.1).
[0146] FIG. 27 exhibits a parallel fold-change plot generated using transcripts related to the photosystem through the ‘Generate Plot’ button (3-12). The plot reveals a consistent pattern: a significant decrease in the expression level (fold-change) of antenna proteins related to the photosystem at 22 DPI across all genotypes (8.1.4, 8.2.4, 8.3.4, 8.4.4, and 8.5.4) than the wildtype (0dpi).System Architectures and Modules
[0147] FIG. 28 illustrates a modular computer-implemented system for pathway-centric interpretation of quantitative biological measurement data. A statistical pipeline module (2810) processes biological measurement data to compute fold-change and significance metrics across multiple experimental comparison conditions. A pathway selection and enrichment module (2820) identifies candidate biological pathways for analysis. A parallel fold-change plot generation module (2830) generates structured pathwaylevel representations of biological entity regulation.
[0148] A pathway-centric context builder (2840) organizes pathway structure and quantitative data into a structured analytical framework. In parallel, a biological database validation module (2850) verifies identifiers and retrieves pathway annotations, while a multi-omics integration module (2860) incorporates proteomic, transcriptomic, or other omics data.
[0149] An inference engine (2870) generates pathway-level analytical outputs, wherein in one embodiment the inference engine (2870) comprises at least one artificial intelligence model configured to process a structured pathway data object and may be instantiated as a local executable, a containerized service, or a remote API endpoint to support flexible deployment including local execution or cloud-based processing. In certain embodiments, the inference engine comprises transformer-based models, including large language models (LLMs) and / or vision-language models (VLMs), capable of performing structured biological reasoning using quantitative, relational, and metadata inputs. The analytical outputs are assessed for robustness by a cross-validation module (2880). Final validated results can be provided through an interpretation output module (2890) in a user-accessible format.
[0150] FIG. 29 illustrates a multimodal user interface for generating pathway-centric interpretations. When the parallel fold-change plot is initially generated, it is displayed in a docked visualization window- - 24 - 187285220.1(2910), and an AI interpretation window (3-17) appears adjacent to the docked visualization window to display pathway-level analytical output generated by an artificial intelligence inference engine. The docked visualization window (2910) can display a parallel fold-change plot (2902) generated by the parallel foldchange plot generation module (2830) as illustrated in FIG. 35. Through a selection interface module (2920), the user may provide input via either: a text-based path (2930), by selecting feature profiles to extract associated quantitative data: or an image-based path (2940), by selecting a region of a pathway image, which is processed to identify and standardize biological entities. In certain embodiments, the image-based path processes user-selected regions of a pathway image using a vision-language model (VLM) to derive spatial and contextual biological insights from the selected region.
[0151] In both cases, structured data are transmitted to a pathway-centric context builder (2840), and an inference engine invocation panel (2911) enables execution of pathway-level interpretation by invoking the inference engine (2870).
[0152] FIG. 30 illustrates a pathway-centric context builder (2840) configured to generate a structured input for an inference engine (2870). The context builder comprises: a pathway identifier component (2841) that stores identification data for a selected biological pathway; a metabolite data component (2842) that organizes biological entities associated with the selected pathway; a statistical value component (2843) that associates quantitative metrics with each biological entity; and an experimental metadata component (2844) that incorporates structured experimental information.
[0153] These components can be integrated into a structured context output (2845), representing a unified pathway-centered data structure. In one embodiment, tire structured output is formatted as a machine-readable context document that includes organism-specific pathway information and validated compound-enzyme-reaction relationships. The context builder implements a validation report component that verifies biological entity identifiers and applies a dynamic fetch limit policy to optimize the size of the structured output for inference models. The structured context output is transmitted to the inference engine (2870) for pathway-level interpretation. The structured context output can integrate: (i) a selected pathway identifier; (ii) validated biological entity identifiers mapped to standardized pathway entities; (iii) quantitative statistical metrics associated with each entity across multiple experimental conditions; and (iv) experimental metadata including genotype, treatment, and comparison context.
[0154] FIG. 31 illustrates a biological database validation module (2850) configured to verify and enrich pathway-related information prior to interpretation. The biological database validation module can include an identifier mapping component (2851) that converts biological entity labels into standardized database identifiers (e.g.. KEGG). A pathway relationship verification component (2852) confirms structural association of the identified entities with a selected biological pathway. A context expansion component (2853) retrieves additional relational or structural data from the database to prepare verified reference- 25 - 187285220.1information. The relational information retrieved may include reaction connectivity, structural organization, and enzyme-substrate relationships.
[0155] The verified and enriched information is transmitted to the pathway-centric context builder (2840), thereby constraining interpretation to validated biological entities and enhancing pathway-level context prior to inference.
[0156] FIG. 32 illustrates a cross-validation module (2880) configured to enhance interpretative reliability by comparing outputs from multiple inference processes and validating them against authoritative biological data.
[0157] An inference engine (2870) generates pathway-level interpretations from a structured context. The cross-validation module includes a primary inference component (2881). a verification component (2882) configured to perform constraint-based fact checking, and a summary component (2883) configured to consolidate the verified interpretation. The verification component is configured to compare an initial draft interpretation generated by the primary inference component against authoritative biological database records retrieved from the biological database validation module. The summary component then aggregates the validated facts and hypotheses into a unified output format prior to evaluation by an output comparison component (2884).
[0158] The verification process applies mandatory constraints to ensure biological accuracy, including exclusion of external knowledge by restricting the generation of compounds, reactions, or pathways not explicitly present in the provided database records; evidence-based rewriting by correcting or removing statements in the draft that contradict the database records; and hypothesis marking by explicitly labeling statements that are logically inferable but not explicitly supported by the database. Furthermore, the verification process can include deterministic inference control by configuring the verification model with a temperature parameter set below a specific threshold (e.g., 0.25) to minimize stochastic variation during the validation process.
[0159] An output comparison component (2884) evaluates consistency between the draft interpretation and the verified version. If a parsing error or lack of response occurs during verification, a fallback mechanism can maintain and present the original interpretation to ensure service continuity. A validated interpretation is then transmitted to the interpretation output module (2890).
[0160] FIG. 33 illustrates a multi-omics integration and correlation analysis module (2860) configured to incorporate structured experimental information into a pathway-centric interpretation framework. The module can include a genotype parsing component (2861), a treatment parsing component (2862), and an experimental parameter structuring component (2863), each configured to extract and organize experimental design information.
[0161] The module further includes a multi-omics correlation component (2864) configured to compute - 26 - 187285220.1correlation coefficients between biological entity measurements and complementary omics-derived data, such as proteomic or transcriptomic profiles. These correlation metrics are integrated into the structured metadata to enable cross-layer validation and mechanistic reasoning within the inference engine (2870). In certain embodiments, metabolite-level quantitative change values are integrated with complementary omics-derived information, including but not limited to proteomics and transcriptomics profiles, aligned to standardized pathway entities within the pathway-centric analytical framework to support structured, AI-assisted biological reasoning.
[0162] The biological data visualization system can be further configured to apply a dynamic data fetch limit policy for retrieving information from an external biological database based on the number of compounds in the input. In one embodiment, the fetch limit is inversely proportional to the compound count. For example, when the compound count is relatively low (e.g., less than or equal to 5), a higher fetch limit (e.g., 15) may be applied; whereas when the count is relatively high (e.g., exceeds 20), a lower fetch limit (e.g., 5) may be applied to ensure optimal context size for artificial intelligence inference. The resulting structured metadata output (2865) is transmitted to the pathway-centric context builder (2840), thereby integrating multi-omics correlation data and experimental context into the pathway-level interpretation process.
[0163] FIG. 34 illustrates an interpretation output module (2890) configured to present pathway-level results generated by an inference engine (2870). The module can include a pathway-level interpretation generation component (2891) that organizes interpretative content with respect to a selected biological pathway, and an evidence / consistency formatting component (2892) that aligns interpretative statements with associated quantitative data. The interpretation output module (2890) may further include a follow-up interpretation execution component (2894) configured to generate and execute one or more follow-up prompts based on generated interpretation results. The follow-up interpretation execution component (2894) interact with the inference engine (2870) to perform additional inference operations, contextual refinement, iterative analysis, evidence validation, related pathway analysis, or secondary interpretation generation associated with the biological pathway (see FIG. 44).
[0164] The biological data visualization system can be configured to embed explicit instructions within the structured context provided to the inference engine. These instructions require the model to cite evidence from the provided context when making biochemical claims and to refrain from generating information not present in the validated pathway data. This can prevent hallucination and help ensure biologically constrained reasoning.
[0165] In one embodiment, the formatted interpretation includes specific pathway-level insights derived from quantitative fold-change variations, such as pathway activation states, (e.g.. "‘Glycolysis pathway is activated under Drug A treatment based on significant positive fold-change values”), pathway inhibition- 27 - 187285220.1states, (e.g., “TCA cycle is inhibited due to enzyme downregulation reflected in negative fold-change trends”), and regulatory network states derived from biological entity correlation patterns and fold-change trajectories, (e.g., “Metabolic reprogramming mechanism inferred from metabolite -enzyme correlation patterns and fold-change trajectories”). A formatted interpretation can be provided through an output presentation interface (2893), thereby delivering validated pathway-level results in a user-accessible format.
[0166] FIG. 35 illustrates a parallel fold-change plot generation module (2830) configured to produce a structured visualization of biological entity regulation across experimental conditions.
[0167] The module includes a statistical input interface (2831) that receives quantitative metrics, a pathway-based filtering component (2832) that selects biological entities associated with a chosen pathway, and an experimental condition vector construction component (2833) that generates condition-specific value vectors.
[0168] A clustering component (2834) groups biological entities by regulatory similarity, and a visual attribute assignment and trajectory construction component (2835) renders fold-change profiles. The resulting parallel fold-change plot output (2902) provides a pathway-centered visualization for downstream analysis and user interaction.
[0169] FIG. 36 illustrates a structured pathway-constrained parallel fold-change visualization generated from metabolite data prior to supplying to an artificial intelligence inference engine, comprising: (3610) a biological entity axis limited to metabolites associated with a selected biological pathway; (3620) a plurality of parallel experimental condition axes corresponding to distinct comparison conditions; (3630) for each metabolite, a fold-change trajectory connecting quantitative values across the condition axes forming a multi-dimensional regulation profile: (3640) visual differentiation attributes applied to denote cluster membership or regulatory grouping; and a zero baseline separating a positive fold-change region (3650) from a negative fold-change region (3660). This visualization represents the structured pathway data object that is transformed and supplied to an Al inference engine for pathway-level interpretation.
[0170] For each biological entity, a fold-change trajectory (3630) is generated by connecting quantitative values across the condition axes, thereby forming a multi-dimensional regulation profile. Visual differentiation attributes (3640), including but not limited to one of color, line pattern, or graphical encoding, are consistently applied to denote cluster membership or regulatory grouping. The resulting structure provides a pathway-aligned, multi -condition quantitative representation in which biological entity identity, regulatory magnitude, directionality, and cluster classification are integrated into a unified, selectable visualization suitable for computational interpretation and user interaction.Al Configuration and Interpretation Interface
[0171] FIG. 39 illustrates an example configuration window for an artificial intelligence system. The LLM configuration window (39-3) provides a user interface for selecting and configuring a large language - 28 - 187285220.1model for pathway-level interpretation. A content-type selector (39-1) enables selection of content generation types, including draft generation, review generation, audit generation, summary generation, vision-based generation, and follow-up query generation. A model selector (39-2) displays information regarding a user-selected LLM model and a server on which the selected LLM model is deployed, and allows the user to select or change the LLM model. The model selector (39-2) includes fields for LLM endpoint URL, model architecture, model selection, and model information including model ID, state, publisher, architecture, and model type. A connection control element (39-4) is configured to initiate communication with a server using a server address and connection configuration information, with the connection control element (39-4) displaying a connected status and providing a reconnect option. The LLM configuration window (39-3) further includes options for profiles review, profile update, generation settings, max tokens configuration, cache prompt selection, and non-thinking mode selection, along with apply and cancel buttons for confirming or discarding configuration changes.
[0172] FIG.40 illustrates an example interface for receiving experimental metadata associated with input data. The interface includes an omics layer selection component (40-1) configured to select a specific omics layer including metabolomics, proteomics, and transcriptomics. A metadata input component (40-2) is configured to receive experimental metadata associated with experimental data, including sample collection, preparation, measurement, and transformation metadata across omics layers. An organism selection component (40-3) is configured to select an organism corresponding to the experimental data.
[0173] FIG. 41 illustrates an example interface for selecting a target biological entity category associated with input data. The interface includes a hierarchical tree structure component (41-1) configured to organize biological entity groups in a navigable tree fonnat, displaying categories such as Eukaryotes, Animals, Plants, and subcategories including Eudicots. Monocots. Fungi, Protists, Prokaryotes, Bacteria, and Archaea. A search input component (41-2) is configured to receive a search input for identifying a biological entity associated with experimental data, displaying fields for Code, Organism, and search parameters.
[0174] FIG. 42 illustrates an example user interface for receiving a text-based input for interpreting a selected data visualization using a large language model. The interface includes a system prompt input component (42- 1 ) configured to receive a system prompt for a large language model, with a service selector and copy, paste, clear, save, and delete controls. A user prompt input component (42-2) is configured to receive a prompt for interpreting user data, with a template selector and similar controls. A selected texts component (42-3) is configured to provide one or more preconfigured prompts and receive a user selection thereof. A prompt selection component (42-4) appears adjacent to both the system prompt input component (42-1) and the user prompt input component (42-2), providing dropdown menus for selecting predefined prompts or templates. A validation control element (42-5) is configured to selectively perform error- 29 - 187285220.1checking and validation with respect to KEGG data during interpretation. An execution control component (42-6) includes a run control configured to initiate processing based on configured settings and a cancel control configured to abort the processing.
[0175] FIG. 43 illustrates an example interface for selecting a region from a pathway image based on a parallel fold-change plot and performing Vision LLM-based interpretation. The interface comprises a tool component (43-1) configured to load a specific pathway image in response to a double-click input on a group name displayed in a parallel fold-change plot, and to allow selection of a target region within the pathway image. A preview component (43-2) is configured to display a preview image of a selected region after the target region is selected, wherein the selected region includes one or more selected metabolites. The preview component (43-2) displays a pathway diagram containing multiple enzyme labels which represent enzymes within the selected pathway region. An execution component (43-3) is configured to initiate interpretation based on metabolite names, pathway information, and enzyme names associated with the selected region. A result display component (43-4) is configured to display interpretation results generated by the LLM, showing a table with columns for type, label, KEGG ID, canonical information, and summary data corresponding to the identified enzymes within the selected pathway region.
[0176] FIG. 44 illustrates an example window displaying Al-generated results, a follow-up prompt interface for refining initial analysis results, and a corresponding result display screen. A statistical metrics region (44-1) displays statistical metrics related to rendering of a large language model, including timing information and token counts. A reasoning model output region (44-2) displays output of a reasoning model, presenting model reasoning information including pathway analysis notes and interpretation guidance. A follow-up prompt region (44-3) provides a region for receiving a prompt for follow -up queries with respect to initial analysis results, allowing users to refine or extend the pathway interpretation. A processing status region (44-4) displays a processing status of the large language model and stores results generated by the large language model. An interpretation results region (44-5) displays interpretation results generated by the large language model, presenting a hierarchical pathway map structure that includes experimental conditions, pathway entries with KEGG identifiers, and phenotypic outcomes such as photoprotection, photosynthetic efficiency, metabolic shifts, stress hormone levels, and stress response pathways. A followup queries region (44-6) is configured to generate, by the large language model, recommended follow-up queries for a user, providing suggested next steps for further analysis. An interpretation result storage region (44-7) stores interpretation results in correspondence with query information, generation time information, and revision history information associated therewith, such that the interpretation results are selectively retrievable, reviewable, comparable, and inspectable by a user.
[0177] FIG. 45 is a system architecture diagram illustrating a client device (4502) and a server device (4510) communicatively coupled via a network. The client device (4502) is configured to execute an- 30 - 187285220.1application (4504) that provides a user interface and performs data processing. The server device (4510) includes a GPU-based computing system configured to execute a large language model (LLM) through an LLM inference engine (4512). Tire client device (4502) transmits a request (4506) including input data and / or a prompt to tire server device (4510) through an application programming interface (API). Tire API communication (4514) represents the network-based exchange between the client device (4502) and the server device (4510). In response to receiving the request (4506), the server device (4510) performs inference using the LLM inference engine (4512) and generates a result. The generated result is transmitted back to the client device (4502) as a response (4508), and the client device (4502) displays the result through the application (4504).Example Computing Environment of the Biological Data Visualization System
[0178] FIG. 21 depicts an example of an apparatus 2100 that may be used to implement the various techniques and features described in the present document. The apparatus 2100 includes one or more processors 2102. The one or more processors 2102 may be configured to read from a memory 2106 and implement the data analysis tool and produce an interactive display on the display interface 2104. The display interface 2104 may comprise a computer-monitor or a light emitting diode (LED) or LCD display.
[0179] Figure 37 is a block diagram showing some of the components typically incorporated in at least some of the computer systems 3700 and other devices on which the disclosed system operates in accordance with some implementations of the present technology. As shown, an example computer system 3700 can include: one or more processors 3702, main memory 3706, non-volatile memory 3710, a network interface device 3712, video display device 3718, an input / output device 3720, a control device 3722 (e.g., keyboard and pointing device), a drive unit 3724 that includes a machine-readable medium 3726, and a signal generation device 3730 that are communicatively connected to a bus 3716. The bus 3716 represents one or more physical buses and / or point-to-point connections that are connected by appropriate bridges, adapters, or controllers. Various common components (e.g., cache memory) are omitted from Figure 37 for brevity. Instead, the computer system 3700 is intended to illustrate a hardware device on which components illustrated or described relative to the examples of the figures and any other components described in this specification can be implemented.
[0180] The computer system 3700 can take any suitable physical form. For example, the computer system 3700 can share a similar architecture to that of a server computer, personal computer (PC), tablet computer, mobile telephone, wearable electronic device, network-connected (“smart”) device (e.g., a television or home assistant device), AR / VR systems (e.g., head-mounted display), or any electronic device capable of executing a set of instructions that specify action(s) to be taken by the computer system 3700. In some implementations, the computer system 3700 can be an embedded computer system, a system -on-chip (SOC), a single-board computer system (SBC) or a distributed system such as a mesh of computer- 31 - 187285220.1systems or include one or more cloud components in one or more networks. Where appropriate, one or more computer systems 3700 can perform operations in real time, near real time, or in batch mode.
[0181] Tire network interface device 3712 enables the computer system 3700 to exchange data in a network 3714 with an entity that is external to the computing system 3700 through any communication protocol supported by the computer system 3700 and tire external entity. Examples of tire network interface device 3712 include a network adaptor card, a wireless network interface card, a router, an access point, a wireless router, a switch, a multilayer switch, a protocol converter, a gateway, a bridge, bridge router, a hub, a digital media receiver, and / or a repeater, as well as all wireless elements noted herein.
[0182] The memory (e.g., main memory 3706, non-volatile memory 3710, machine-readable medium 3726) can be local, remote, or distributed. Although shown as a single medium, the machine -readable medium 3726 can include multiple media (e.g., a centralized / distributed database and / or associated caches and servers) that store one or more sets of instructions 3728. The machine-readable (storage) medium 3726 can include any medium that is capable of storing, encoding, or carrying a set of instructions for execution by the computer system 3700. The machine -readable medium 3726 can be non-transitory or comprise a non-transitory device. In this context, a non-transitory storage medium can include a device that is tangible, meaning that the device has a concrete physical form, although the device can change its physical state. Thus, for example, non-transitory refers to a device remaining tangible despite this change in state.
[0183] Although implementations have been described in the context of fully functioning computing devices, the various examples are capable of being distributed as a program product in a variety of forms. Examples of machine-readable storage media, machine-readable media, or computer-readable media include recordable-type media such as volatile and non-volatile memory, removable memory, hard disk drives, optical disks, and transmission-type media such as digital and analog communication links.
[0184] In general, the routines executed to implement examples herein can be implemented as part of an operating system or a specific application, component, program, object, module, or sequence of instructions (collectively referred to as “computer programs”). The computer programs typically comprise one or more instructions (e.g., instructions 3708, 3728) set at various times in various memory and storage devices in computing device(s). When read and executed by the processor 3702, the instruction(s) cause the computer system 3700 to perform operations to execute elements involving the various aspects of the disclosure.
[0185] Figure 38 is a system diagram illustrating an example of a computing environment in which the disclosed system operates in some implementations. In some implementations, environment 3800 includes one or more client computing devices 3805A-D, examples of which can host the biological data visualization system. Client computing devices 3805 operate in a networked environment using logical connections through network 3830 to one or more remote computers, such as a server computing device.
[0186] In some implementations, server computing device 3810 is an edge server which receives client- 32 - 187285220.1requests and coordinates fulfillment of those requests through other servers, such as servers 1120A-C. In some implementations, server computing devices 3810 and 3820 comprise computing systems, such as the biological data visualization system. Though each server computing device 3810 and 3820 is displayed logically as a single server, server computing devices can each be a distributed computing environment encompassing multiple computing devices located at the same or at geographically disparate physical locations. In some implementations, each server computing device 3820 corresponds to a group of servers.
[0187] Client computing devices 3805 and server computing devices 3810 and 3820 can each act as a server or client to other server or client devices. In some implementations, servers (3810, 3820A-C) connect to a corresponding database (3815, 3825A-C). As discussed above, each server computing device 3820 can correspond to a group of servers, and each of these servers can share a database or can have its own database. Databases 3815 and 3825 warehouse (e.g., store) information such as claims data, email data, call transcripts, call logs, policy data and so on. Though databases 3815 and 3825 are displayed logically as single units, databases 3815 and 3825 can each be a distributed computing environment encompassing multiple computing devices, can be located within their corresponding server, or can be located at the same or at geographically disparate physical locations.
[0188] Network 3830 can be a local area network (LAN) or a wide area network (WAN), but can also be other wired or wireless networks. In some implementations, network 3830 is the Internet or some other public or private network. Client computing devices 3805 are connected to network 3830 through a network interface, such as by wired or wireless communication. While the connections between server computing device 3810 and server computing devices 3820 arc shown as separate connections, these connections can be any kind of local, wide area, wired, or wireless network, including network 3830 or a separate public or private network.Example Technical Solutions
[0189] Some example technical solutions implemented by preferred embodiments include the following.
[0190] 1. A method of providing interactive data analysis, perfonned by a computer comprising one or more processors, comprising: receiving an input comprising a plurality of sample groups comprising experimental data, wherein each sample group corresponds to data obtained in an experiment using a particular value of a feature from a plurality of features; calculating fold-change values representing changes among the sample groups; determining a number of labels representative of the fold-change values, wherein each label identifies two sample groups; mapping tire plurality of features along a first direction of a graph; mapping the labels along a second direction of the graph; depicting, for each feature, a corresponding curve, wherein data points of the curve correspond to fold-change values for the feature across the labels; associating each data point with a corresponding pathway map that depicts a pathway: verifying, against a standardized biological database, feature identifiers corresponding to the plurality of - 33 - 187285220.1features; transforming quantitative data represented in the graph into a structured, machine -readable pathway data object integrating the verified feature identifiers and the fold-change values; supplying the structured pathway data object to an artificial intelligence inference engine to generate a pathway-level analytical output; receiving a user selection of one or more specific data points; obtaining, by retrieving from a database, graphical information of a pathway corresponding to the one or more specific data points; validating the pathway-level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the pathway-level analytical output entities or relationships not present in the structured pathway data object; and making an interactive visual presentation of the pathway on a display interface such that the visual presentation identifies features in the pathway, their deviations during the experiment, and the pathway-level analytical output.
[0191] 2. Tire method of example 1. wherein verifying comprises mapping the feature identifiers to standardized database identifiers and confirming structural association of the features with a biological pathway.
[0192] 3. The method of example 1, further comprising: identifying one or more biological pathways associated with the plurality of features; and selecting a biological pathway from the one or more biological pathways, wherein the depicting is limited to features associated with the selected biological pathway.
[0193] 4 The method of example 3, wherein the graph comprises: a feature axis limited to features associated with a selected biological pathway; a plurality of parallel axes corresponding to distinct experimental conditions; and for each feature, a fold-change trajectory connecting data points representing fold-change values across the plurality of condition axes.
[0194] 5. The method of example 1. wherein the structured pathway data object is configured to constrain the artificial intelligence inference engine to the verified feature identifiers.
[0195] 6. The method of example 1, wherein tire at least one constraint comprises at least one of: exclusion of external knowledge by restricting generation of compounds, reactions, or pathways not present in the structured pathway data object; evidence-based rewriting by correcting or removing statements that contradict the structured pathway data object: or hypothesis marking by labeling statements that are logically inferable but not explicitly supported by the structured pathway data object.
[0196] 7. Tire method of example 1, wherein the pathway-level analytical output comprises at least one of a pathway activation state, a pathway inhibition state, or a regulatory’ netw ork state.
[0197] 8. A method of providing interactive data analysis for simultaneous comparison of multiple sample groups, performed by a computer comprising one or more processors, comprising: receiving an input comprising a plurality of sample groups comprising experimental data, wherein each sample group corresponds to data obtained in an experiment using a particular value of a feature from a plurality of features; calculating, for the input, fold-change values representing changes among the sample groups;- 34 - 187285220.1determining a number of labels representative of the fold-change values, wherein each label identifies two sample groups; mapping the plurality of features along a first direction of a graph; mapping tire labels along a second direction of the graph; depicting, for each feature, a corresponding curve connecting data points across the labels to form a multi-condition fold-change trajectory, wherein data points of tire curve correspond to fold-change values for tire feature across the labels for simultaneous visualization of feature regulation across a plurality of sample group comparisons; establishing a link between each data point and a corresponding pathway map that depicts a pathway; receiving a user selection of one or more specific data points; obtaining, by retrieving from a database, graphical information of a pathway corresponding to the one or more specific data points; and making an interactive visual presentation of the pathway on a display interface such that the visual presentation identifies features in the pathway and their deviations during the experiment, wherein the interactive visual presentation is dynamically updated in response to user selection of different data points.
[0198] 9 The method of example 8, wherein the plurality of features corresponds to a plurality of metabolites used in the experiment.
[0199] 10. The method of example 8, wherein the calculating the fold-change values comprises evaluating: fold-change = power (2, Logzfold-change), wherein Log2fold-change = log2 (avg2 / avgl); wherein fold-change represents a fold-change value between a first sample group and a second sample group, power (2, x) is a function that outputs a value of 2x, log2 is logarithm with base 2, avgl represents an average of the first sample group, and avg2 represents an average of the second sample group.
[0200] 11. The method of example 8, wherein the visual presentation visually identifies features based on color-coding, wherein negative fold-change values are displayed in a first color, positive fold-change values are displayed in a second color, and areas with no change are displayed in a third color.
[0201] 12. The method of example 8, w herein tire database is remotely coupled to tire computer through a network connection, and wherein the obtaining comprises: formulating a query using the one or more specific data points; transmitting the query over the network connection to the database; and receiving a response from the database comprising the graphical information of the pathway.
[0202] 13. The method of example 8, further comprising: determining, at least partly based on a user selection, a database from which to retrieve the graphical information of the pathway.
[0203] 14. Tire method of example 8, further comprising: selecting features using a category-based approach, wherein the category-based approach comprises selecting features based on a main category and a sub-category, and generating the graph using the selected features.
[0204] 15. A system for generating pathway-level analytical outputs from biological measurement data, comprising: one or more processors; and a memory storing instructions that, when executed by the one or more processors, cause the system to: receive quantitative measurement data for a plurality of biological - 35 - 187285220.1entities across a plurality of experimental conditions; compute fold-change values for each biological entity across the plurality of experimental conditions; identify one or more biological pathways associated with the biological entities; verily, against a standardized biological database, biological entity identifiers corresponding to the biological entities; transform the fold-change values into a structured, machine-readable pathway data object integrating the verified biological entity identifiers and the fold-change values; supply the structured pathway data object to an artificial intelligence inference engine to generate a pathway-level analytical output; and validate the pathway-level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the pathway-level analytical output entities or relationships not present in the structured pathway data object.
[0205] 16. The system of example 15, wherein each experimental condition corresponds to at least one of a distinct genotype, treatment, or experimental parameter, wherein the experimental parameter comprises at least one of a time point, dosage level, environmental conditions, sample type, or tissue source.
[0206] 17. The system of example 15, wherein the pathway data object comprising at least one of: a pathway identifier for a biological pathway of the one or more biological pathways, the verified biological entity identifiers mapped to standardized database entries, or the fold-change values for each biological entity across the plurality of experimental conditions.
[0207] 18. The system of example 15, wherein the at least one constraint comprises at least one of: exclusion of external knowledge by restricting generation of compounds, reactions, or pathways not present in the structured pathway data object;evidence-based rewriting by correcting or removing statements that contradict the structured pathway data object; orhypothesis marking by labeling statements that are logically inferable but not explicitly supported by the structured pathway data object.
[0208] 19. The system of example 15, wherein the memory further stores instructions that cause the system to: generate a pathway-constrained visualization depicting, for each biological entity associated with a selected biological pathway of the one or more biological pathways, a curve connecting fold-change values across the plurality of experimental conditions; and establish a link betw een each data point of the curve and a corresponding pathway map.
[0209] 20. The system of example 15, wherein the memory further stores instructions that cause the system to: generate a visualization depicting, for each biological entity, a curve connecting fold-change values across the plurality of experimental conditions; and in response to a user selection of a data point, display a data table comprising at least one of a log2 fold-change value, a fold-change value, a p-value, or an adjusted p-value for the biological entity corresponding to the selected data point.
[0210] 21. A computer-implemented method for validating artificial intelligence-generated biological pathway interpretations, comprising: generating a parallel fold-change plot from quantitative biological measurement data; transforming quantitative data from the parallel fold-change plot into a structured,- 36 - 187285220.1machine-readable pathway data object; inputting the structured pathway data object to an artificial intelligence inference engine to generate an initial pathway-level analytical output comprising natural language text; validating the initial pathway -level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the initial pathway-level analytical output entities or relationships not present in the structured pathway data object; verifying biological database identifiers and related information in the initial pathway-level analytical output by comparing against a standardized biological database, thereby identifying discrepancies between the initial pathwaylevel analytical output and records of the standardized biological database; supplying corrected or validated data to the artificial intelligence inference engine to generate a revised pathway-level analytical output, wherein tire corrected or validated data is based on at least one of the validation or verification results; presenting the revised pathway-level analytical output alongside the parallel fold-change plot in a synchronized side-by-side display interface, wherein selection of a data point in the parallel fold-change plot highlights a corresponding statement in the revised pathway-level analytical output, and wherein selection of a statement in the revised pathway-level analytical output highlights a corresponding data point in the parallel fold-change plot; generating one or more suggested follow-up queries based on the revised pathway-level analytical output; and receiving user input comprising at least one of: a follow-up query based on tire one or more suggested follow-up queries, or a modification to at least one of the one or more suggested follow-up queries.
[0211] 22. The method of example 21, wherein validating the initial pathway-level analytical object comprises: excluding generation of compounds, reactions, or pathways not present in the structured pathway data object; correcting or removing statements that contradict the structured pathway data object; and labeling statements that are logically inferable but not explicitly supported by the structured pathwaydata object.
[0212] 23. The method of example 21, wherein verifying biological database identifiers comprises: mapping the biological database identifiers to standardized database identifiers; confirming structural association of biological entities identified by tire biological database identifiers with a selected biological pathway; and retrieving additional relational or structural data from the standardized biological database for preparing verified reference information.
[0213] 24. The method of example 21, wherein generating the one or more suggested follow-up queries comprises: analyzing content of the revised pathway -level analytical output; generating the one or more suggested follow-up queries based on identified pathways, biological entities, and regulatory- states identified in the revised pathway-level analytical output: and presenting the one or more suggested followup queries to a user for selection or modification.
[0214] 25. The method of example 21, wherein the biological database identifiers comprise KEGG- 37 - 187285220.1identifiers, and wherein the standardized biological database comprises a KEGG database.
[0215] 26. The method of example 1, further comprising: receiving user input via at least one of a text-based input path or an image-based input path, wherein tire text-based input path comprises selecting feature profiles to extract associated quantitative data, and wherein the image-based input path comprises selecting a region of a pathway image and processing the selected region to identify and standardize biological entities depicted within the selected region.
[0216] 27. The method of example 1, wherein the curve for each feature connects data points across the labels to form a multi-condition fold-change trajectory, enabling simultaneous visualization of feature regulation across a plurality of sample group comparisons.
[0217] A computer platform comprising one or more processors or one or more processor cores configured to read instructions from a memory and implement an above described method.
[0218] A computer-readable medium having code stored thereon, the code, upon execution by one or more processors, causing the one or more processors to implement an above-described method.Final Remarks
[0219] The Volcano Plot is an invaluable tool for visualizing the differences and statistical significance of features when comparing two groups, greatly aiding in analysis. However, analyzing feature changes and statistical significance across multiple groups for comparative analysis can be a daunting task. This patent document introduces the parallel fold-change plot, which efficiently identifies patterns of feature changes among multiple groups. With the algorithm that generates the parallel fold-change plot, users can analyze data flexibly by selecting the number of comparison groups and the number of features of interest in large datasets or big data. It also greatly reduces the time required for data analysis and interpretation, enabling faster feedback for decision-making in the next experiment. It can be applied in the following fields:
[0220] Parallel fold-change plot applies to multiple comparisons of any sample groups in any sample type using fold-change and p-value, which represents the level of marginal significance within a statistical hypothesis test.
[0221] Additionally, it can be applied to the discovery’ of compounds on a large scale, in complex mixture sample analysis such as crude oil analysis.Example Technical Benefits
[0222] Currently, the research and development cycle involving experimentation, data analysis, and subsequent experimentation based on data analysis is a prolonged process, often spanning several days due to the bottleneck of data analysis. The techniques outlined in this document streamline this cycle by¬ providing a swift analysis of experimental results, enabling the rapid identification of distinct metabolites- 38 - 187285220.1and their quantitative variations within experiment-utilized pathways. For instance, following the collection of experimental data, a researcher can, within a few hours, assess the influence of specific features or metabolites on the results. This expedited analysis empowers researchers to promptly adjust and refine subsequent experiments based on insights gleaned from the initial data, significantly accelerating the research and development cycle.
[0223] Beyond accelerating the data analysis cycle, the parallel fold-change visualization enables simultaneous comparison of biological entity regulation across dozens or hundreds of experimental conditions in a single view. Rather than manually inspecting individual pairwise comparisons, each of which would conventionally require a separate volcano plot, researchers can identify coordinated crosscondition regulatory patterns from a unified visualization. For example, an experiment comparing 8 genotypes across 3 treatments can generate a substantial number of individual pairwise comparisons, all of which are consolidated into a single parallel fold-change plot. The parallel fold-change visualization framework encodes biological entity identity, multi-condition quantitative vectors, regulatory directionality, and cluster membership into a unified graphical framework that is both human-interpretable and computationally selectable. This visualization is not merely a static display of data, but a specific technical architecture for organizing and representing biological data that enables downstream computational processing.
[0224] The transformation of visualization-aligned quantitative data into structured, machine-readable pathway data objects further enables automated pathway-level interpretation that is both reproducible and biologically constrained. By integrating pathway identifiers, validated biological entity identifiers, quantitative statistical metrics, and experimental metadata into a unified data structure, the system provides a standardized input for artificial intelligence inference engines. The resulting pathway-level analytical outputs, including pathway activation states, pathway inhibition states, and regulatory network states, are derived from verified biological relationships rather than unconstrained model generation. The novelty of the invention resides in this transformation pipeline and the structured data objects it produces, with Al models serving as one processing component that receives and processes these novel inputs.
[0225] Biological accuracy is maintained through a database validation module that verifies biological entity identifiers against standardized databases and confirms structural association of entities with selected biological pathways prior to inference. More specifically, the validation module can (1) convert biological entity labels into standardized database identifiers, (2) confirm structural association of identified entities with selected biological pathways, and (3) retrieve additional relational or structural data to prepare verified reference information. The cross-validation module further applies mandatory constraints, including exclusion of external knowledge, evidence -based rewriting, and hypothesis marking, to ensure that AI-generated interpretations are grounded in verified biological facts. The invention can help prevent- 39 - 187285220.1hallucinations by requiring the inference model to cite evidence from the provided context and restrict it from generating information beyond validated pathway data. Deterministic inference control, configured via a temperature parameter set below a specific threshold, can minimize stochastic variation during the validation process, improving reproducibility of pathway-level analytical outputs.
[0226] Cross-layer validation and mechanistic reasoning are supported through a multi-omics integration module that computes correlation coefficients between biological entity measurements and complementary' omics-derived data, such as proteomic or transcriptomic profiles. By aligning these measurements to standardized pathway entities within the pathway-centric analytical framework, the system incorporates multiple data modalities into a single structured interpretation context.
[0227] Researchers can interact with the system through both text-based selection of feature profiles and image-based selection of pathway regions. In certain embodiments, the image-based path processes user-selected regions of a pathway image using a vision-language model to derive spatial and contextual biological insights, enabling intuitive construction of interpretation contexts directly from pathway visualizations.
[0228] Context size for inference models is managed through a dynamic data fetch limit policy that adjusts tire volume of retrieved biological database information based on the number of compounds in the input. When the compound count is relatively low, (e.g., less than or equal to 5), a higher fetch limit (e.g., 15) is applied to provide richer context; when the count is relatively' high, (e.g., exceeds 20), a lower fetch limit (e.g., 5) is applied to maintain efficient processing by the inference engine. This reflects a specific technical optimization that ensures optimal context size for artificial intelligence inference.
[0229] Flexible deployment is supported through instantiation of the inference engine as a local executable, a containerized service, or a remote API endpoint, accommodating varying computational resources and institutional infrastructure requirements.
[0230] Tire interactive visual presentation of the pathway on tire display interface may be dynamically updated in response to user selection of different data points. In certain embodiments, the visual presentation identifies features in the pathway, their deviations during the experiment, and tire pathwaylevel analytical output. In response to a user selection of a data point, the system may display a data table comprising at least one of a log2 fold-change value, a fold-change value, a p-value, or an adjusted p-value for the biological entity corresponding to the selected data point.
[0231] The system additionally supports multiple omics workflows. In certain embodiments, a metabolomics targeted workflow supports targeted metabolomics analysis when KEGG IDs are available. In other embodiments, a transcriptomics workflow supports transcriptomics analysis. The system may also support category -based feature selection, wherein features are selected based on a main category and a subcategory', providing an efficient approach to generate a parallel fold-change plot with numerous features.-40 - 187285220.1
[0232] Furthermore, the hybrid method combining automated Al analysis with human expert verification provides enhanced reliability compared to fully automated approaches. By enabling researchers to verify accuracy against original quantitative data and approve final interpretations before publication, error rates can be significantly reduced while maintaining the efficiency benefits of Al-assisted analysis.
[0233] In some implementations, the multiple workflows may be assigned individual processors of a multi-processor platform or individual threads of a multi-thread software implementation, allowing for a more uniform and better utilization of resources of a computer platform. Such an implementation of interactive data analysis, which may be considered a uniform load implementation, provides for responsiveness and accuracy concurrently for multiple data mining instances running in parallel, thus improving overall operation of a computer platform for data mining of biological experimental data.
[0234] While this patent document contains many specifics, these should not be construed as limitations on the scope of any invention or of what may be claimed, but rather as descriptions of features that may be specific to particular embodiments of particular inventions. Certain features that are described in this patent document in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.
[0235] Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. Moreover, the separation of various system components in the embodiments described in this patent document should not be understood as requiring such separation in all embodiments.
[0236] Only a few implementations and examples are described and other implementations, enhancements and variations can be made based on what is described and illustrated in this patent document.- 41 - 187285220.1
Claims
CLAIMS1. A method of providing interactive data analysis, performed by a computer comprising one or more processors, comprising:receiving an input comprising a plurality of sample groups comprising experimental data, wherein each sample group corresponds to data obtained in an experiment using a particular value of a feature from a plurality of features;calculating fold-change values representing changes among the sample groups; determining a number of labels representative of the fold-change values, wherein each label identifies two sample groups;mapping the plurality of features along a first direction of a graph:mapping the labels along a second direction of the graph;depicting, for each feature, a corresponding curve, wherein data points of the curve correspond to fold-change values for the feature across the labels;associating each data point with a corresponding pathway map that depicts a pathway; verifying, against a standardized biological database, feature identifiers corresponding to the plurality of features;transforming quantitative data represented in the graph into a structured, machine-readable pathway data object integrating the verified feature identifiers and the fold-change values;supplying the structured pathway data object to an artificial intelligence inference engine to generate a pathway-level analytical output;receiving a user selection of one or more specific data points;obtaining, by retrieving from a database, graphical information of a pathway corresponding to the one or more specific data points;validating the pathway-level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the pathway-level analytical output entities or relationships not present in the structured pathway data object; andmaking an interactive visual presentation of the pathway on a display interface such that the visual presentation identifies features in the pathway, their deviations during the experiment, and the pathway-level analytical output.
2. The method of claim 1, wherein verifying comprises mapping the feature identifiers to standardized database identifiers and confirming structural association of the features with a biological pathway.-42 - 187285220.
13. The method of claim 1, further comprising:identifying one or more biological pathways associated with the plurality of features; and selecting a biological pathway from the one or more biological pathways, wherein the depicting is limited to features associated with the selected biological pathway.
4. The method of claim 3, wherein the graph comprises:a feature axis limited to features associated with a selected biological pathway:a plurality of parallel axes corresponding to distinct experimental conditions; andfor each feature, a fold-change trajectory connecting data points representing fold-change values across the plurality of condition axes.
5. The method of claim 1, wherein the structured pathway data object is configured to constrain the artificial intelligence inference engine to the verified feature identifiers.
6. The method of claim 1, wherein the at least one constraint comprises at least one of:exclusion of external know ledge by restricting generation of compounds, reactions, or pathways not present in the structured pathway data object;evidence-based rewriting by correcting or removing statements that contradict the structured pathway data object; orhypothesis marking by labeling statements that are logically inferable but not explicitly supported by the structured pathway data object.
7. The method of claim 1, wherein the pathway-level analytical output comprises at least one of a pathway activation state, a pathway inhibition state, or a regulatory network state.
8. The method of claim 1, further comprising:receiving user input via at least one of a text-based input path or an image-based input path, wherein the text-based input path comprises selecting feature profiles to extract associated quantitative data, and wherein the image-based input path comprises selecting a region of a pathway image and processing the selected region to identify and standardize biological entities depicted within the selected region.
9. The method of claim 1, wherein the curve for each feature connects data points across the labels to form a multi-condition fold-change trajectory, enabling simultaneous visualization of feature regulation across a plurality of sample group comparisons.- 43 - 187285220.
110. A system for generating pathway-level analytical outputs from biological measurement data, comprising:one or more processors; anda memory storing instructions that, when executed by the one or more processors, cause the system to:receive quantitative measurement data for a plurality of biological entities across a plurality of experimental conditions;compute fold-change values for each biological entity across the plurality of experimental conditions;identify one or more biological pathways associated with the biological entities;verify, against a standardized biological database, biological entity identifiers corresponding to the biological entities;transform the fold-change values into a structured, machine -readable pathway data object integrating the verified biological entity identifiers and the fold-change values;supply the structured pathway data object to an artificial intelligence inference engine to generate a pathway-level analytical output; andvalidate the pathway-level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the pathway-level analytical output entities or relationships not present in the structured pathway data object.
11. The system of claim 10, wherein each experimental condition corresponds to at least one of a distinct genotype, treatment, or experimental parameter, wherein the experimental parameter comprises at least one of a time point, dosage level, environmental conditions, sample type, or tissue source.
12. The system of claim 10, wherein the pathway data object comprising at least one of: a pathway identifier for a biological pathway of the one or more biological pathways, the verified biological entity identifiers mapped to standardized database entries, or the fold-change values for each biological entity across the plurality of experimental conditions.
13. The system of claim 10, wherein the at least one constraint comprises at least one of:exclusion of external knowledge by restricting generation of compounds, reactions, or pathways not present in the structured pathway data object;evidence-based rewriting by correcting or removing statements that contradict the structured- 44 - 187285220.1pathway data object; orhypothesis marking by labeling statements that are logically inferable but not explicitly supported by the structured pathway data object.
14. The system of claim 10, wherein the memory further stores instructions that cause the system to: generate a pathway-constrained visualization depicting, for each biological entity associated with a selected biological pathway of the one or more biological pathways, a curve connecting fold-change values across the plurality of experimental conditions; andestablish a link between each data point of the curve and a corresponding pathway map.
15. The system of claim 10, wherein the memory further stores instructions that cause the system to: generate a visualization depicting, for each biological entity, a curve connecting fold-change values across the plurality of experimental conditions; andin response to a user selection of a data point, display a data table comprising at least one of a log2 fold-change value, a fold-change value, a p-value, or an adjusted p-value for the biological entity corresponding to the selected data point.
16. A computer-implemented method for validating artificial intelligence-generated biological pathway interpretations, comprising:generating a parallel fold-change plot from quantitative biological measurement data; transforming quantitative data from the parallel fold-change plot into a structured, machine-readable pathway data object;inputting the structured pathway data object to an artificial intelligence inference engine to generate an initial pathway-level analytical output comprising natural language text:validating the initial pathway-level analytical output by applying at least one constraint that restricts the artificial intelligence inference engine from including in the initial pathway-level analytical output entities or relationships not present in the structured pathway data object;verifying biological database identifiers and related information in the initial pathway-level analytical output by comparing against a standardized biological database, thereby identifying discrepancies between the initial pathway-level analytical output and records of the standardized biological database;supplying corrected or validated data to the artificial intelligence inference engine to generate a revised pathway-level analytical output, wherein the corrected or validated data is based on at least one of the validation or verification results;- 45 - 187285220.1presenting the revised pathway-level analytical output alongside the parallel fold-change plot in a synchronized side-by-side display interface, wherein selection of a data point in the parallel foldchange plot highlights a corresponding statement in the revised pathway-level analytical output, and wherein selection of a statement in the revised pathway-level analytical output highlights a corresponding data point in the parallel fold-change plot;generating one or more suggested follow-up queries based on the revised pathway-level analytical output: andreceiving user input comprising at least one of:a follow-up query based on the one or more suggested follow-up queries, ora modification to at least one of the one or more suggested follow-up queries.
17. The method of claim 16. wherein validating the initial pathway-level analytical object comprises:excluding generation of compounds, reactions, or pathways not present in the structured pathway data object;correcting or removing statements that contradict the structured pathway data object; and labeling statements that are logically inferable but not explicitly supported by the structured pathway data object.
18. The method of claim 16, wherein verifying biological database identifiers comprises:mapping the biological database identifiers to standardized database identifiers; confirming structural association of biological entities identified by the biological database identifiers with a selected biological pathway; andretrieving additional relational or structural data from the standardized biological database for preparing verified reference information.
19. The method of claim 16, wherein generating the one or more suggested follow-up queries comprises:analyzing content of the revised pathway-level analytical output;generating the one or more suggested follow-up queries based on identified pathways, biological entities, and regulatory states identified in the revised pathway-level analytical output; and presenting the one or more suggested follow-up queries to a user for selection or modification.
20. The method of claim 16, wherein the biological database identifiers comprise KEGG identifiers, and wherein the standardized biological database comprises a KEGG database.- 46 - 187285220.1
