An intelligent self-adaptive rasterization method for genome-wide association analysis and molecular phenotype QTL data visualization and a visualization system thereof

Abstract

The present application relates to the field of large-scale data visualization and intelligent sampling technology, and more particularly to a kind of intelligent adaptive rasterization genomic association analysis and molecular phenotype QTL data visualization method and visualization system thereof, the method first carries out spatial gridding division to input high-density data point, data point is mapped to preset grid unit;Subsequently, statistical characteristic information is calculated in each grid unit, including extreme value, point density and attribute aggregation result;On this basis, significance judgment, high-density judgment and random sampling judgment are sequentially executed by decision control module to determine whether to keep the representative point of the corresponding grid unit.The present application can significantly reduce the number of data points and computing overhead while effectively retaining key significant signals and high-density area features, avoiding the loss of important information, and is suitable for genomic association analysis, QTL data display and other efficient visualization and analysis scenarios of large-scale high-dimensional data.

Description

Technical Field

[0001] This invention relates to the fields of bioinformatics and computer graphics, and in particular to an intelligent adaptive rasterization method and visualization system for genome association analysis and molecular phenotype QTL data. Background Technology

[0002] With the development of high-throughput sequencing technology, population genetic analysis such as GWAS and eQTL is widely used, generating massive amounts of association analysis data. Manhattan plots, as a standardized visualization method, are widely used. However, existing plotting techniques have revealed five core problems due to the increased data scale and analytical complexity: First, point-by-point plotting is inefficient and cannot meet the high-throughput batch processing requirements of tens of millions of data points and multi-phenotype analysis. Second, it requires memory to store and render all data points simultaneously, resulting in high memory consumption, potential overflow, and limitations on multi-task parallel plotting. Third, it adopts a fixed plotting strategy without adaptive optimization mechanisms, making it difficult to balance plotting performance with the fidelity of significant association signals. Fourth, most tools are designed for traditional GWAS scenarios and lack efficient support for large-scale, multi-visual-element display of molecular phenotype QTL data. Fifth, although rasterization technology can theoretically alleviate performance and memory issues, the aggregation strategies of existing methods are simple and do not incorporate optimization based on genomic data features, failing to balance signal preservation and data compression, and thus failing to meet the comprehensive visualization requirements of this field. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent adaptive rasterization method and visualization system for genome association analysis and molecular phenotype QTL data, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, this invention provides an intelligent adaptive rasterization method for genome association analysis and visualization of molecular phenotype QTL data, comprising the following steps:

[0005] S1. Obtain association data generated by genome association analysis or molecular phenotypic quantitative trait locus analysis, wherein the association data includes at least genomic location coordinates and association significance indicators;

[0006] S2. Construct a regular grid for data mapping within the target display space, and map the associated data to the corresponding grid cells;

[0007] S3. Based on the numerical and spatial distribution characteristics of the associated data, the data within the grid cell is aggregated to generate at least one representative value;

[0008] S4. Based on the representative value, determine whether to output the corresponding grid cell and generate output data for visualization.

[0009] The aggregation process employs at least a maximum value aggregation strategy to preserve significant correlation signals.

[0010] Furthermore, in step S3, the aggregation process includes multi-level rasterization processing, specifically: the associated data is divided into a set of salient data points and a set of background data points according to a preset saliency threshold, and rasterization processing is performed using grids of different resolutions respectively.

[0011] Furthermore, the salient data point set is rasterized using a first grid resolution, while the background data point set is rasterized using a second grid resolution lower than the first grid resolution.

[0012] Furthermore, for each data point within a grid cell, the maximum correlation significance value is used as the representative value, instead of calculating the average value of the data points.

[0013] Furthermore, in step S4, whether to output a grid cell is determined based on at least one or a combination of the following rules: when the maximum correlation significance value within the grid cell exceeds a significance threshold, the grid cell is output; when the number of data points within the grid cell exceeds a density threshold, the grid cell is output; when the number of data points within the grid cell does not exceed the density threshold, the grid cell is randomly sampled and output.

[0014] Furthermore, after constructing the regular grid, the data point density within each grid cell is first calculated. When the data point density of a grid cell exceeds a preset threshold, the grid cell is subdivided to form multiple sub-grid cells.

[0015] Furthermore, the data points within the subdivided subgrid cells are aggregated separately, while the unsubdivided grid cells retain the original aggregation method.

[0016] Furthermore, when the associated data contains continuous effect values ​​and is represented in a color gradient manner, the continuous effect values ​​are discretized into multiple color intervals, and the data in each color interval is rasterized separately. Then, the rasterization results of each color interval are merged for visualization.

[0017] Furthermore, the visualization system applied to the aforementioned intelligent adaptive rasterization genome association analysis and molecular phenotype QTL data visualization method includes: a data acquisition module for acquiring association analysis data; a grid construction module for constructing a regular grid and mapping data to grid cells; an aggregation processing module for aggregating the data within the grid cells to generate representative values; and an output control module for determining whether to output the corresponding grid cells based on the representative values ​​and generating visualization results; wherein the aggregation processing module employs at least a maximum value aggregation strategy.

[0018] Furthermore, the aggregation processing module is configured to stratify the data according to a significance threshold and process the data at different levels using grids of different resolutions. The system also includes a density analysis module for analyzing the density of data points within the grid cells and triggering grid subdivision processing when the density exceeds a preset threshold. The system is configured to retain color gradient information for visualization when processing molecular phenotypic quantitative trait locus data containing continuous effect values.

[0019] Compared with existing technologies, this invention has at least the following advantages: it can complete the visualization processing of tens of millions or even larger amounts of data within seconds; it significantly reduces memory consumption and avoids memory overflow in multi-chromosome and multi-phenotype plotting scenarios; it preserves the height and morphology of significant association peaks completely, with a false negative rate of less than 1%; it can automatically select the optimal rasterization strategy according to the data scale, chromosome number, and density distribution; it is also applicable to the visualization analysis of traditional GWAS data and molecular phenotype QTL data, and has been developed in three versions: R language package, cloud platform, and command line. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the multi-level rasterization principle of the present invention;

[0021] Figure 2 This is a diagram illustrating the results of genome-wide association analysis of duck plasma cholinesterase based on a multi-level rasterization method according to the present invention.

[0022] Figure 3 This is a flowchart of the intelligent rasterization decision-making process of the present invention;

[0023] Figure 4 This is a diagram illustrating the genome-wide association analysis results of betaine abundance in bovine muscle based on the intelligent rasterization method of this invention;

[0024] Figure 5 This is a schematic diagram of the adaptive density rasterization of the present invention;

[0025] Figure 6 This invention presents a diagram illustrating the cis-cis-eQTL results of duck cecal tissue plotted using an adaptive density rasterization method.

[0026] Figure 7 This is a diagram illustrating the genome-wide association analysis results based on the multi-level rasterization method of this invention;

[0027] Figure 8 This invention presents a diagram illustrating the trans-tran-eQTL results of duck cecal tissue plotted using an adaptive density rasterization method. Detailed Implementation

[0028] The technical solution of the present invention will be further described below with reference to specific embodiments. Those skilled in the art should understand that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Any modifications or equivalent substitutions made to the present invention without departing from its spirit and substance should fall within the scope of protection of the present invention.

[0029] Example 1: A Multi-Level Rasterization Method for Visualizing Genome Association Data

[0030] This embodiment provides an intelligent adaptive rasterization method for genome association analysis and molecular phenotype QTL data visualization, applicable to large-scale association analysis results generated by GWAS and molecular phenotype QTL. A detailed flowchart can be found below. Figure 1 .

[0031] Data Acquisition and Preprocessing: First, acquire the raw data obtained from genome association analysis or molecular phenotype QTL analysis. This data should include at least: genomic location coordinates (chromosome number and physical location); association significance indicators (P-value or its transformed value, such as...). ); optional effect values ​​or color coding information. The data undergoes necessary preprocessing, including missing value filtering and coordinate normalization, to adapt it to the target display space.

[0032] Significance stratification: setting a significance threshold The input data is divided into a salient data point set and a background data point set. The salient data point set includes data points with a p-value less than the saliency threshold, while the background data point set includes data points with a p-value greater than or equal to the saliency threshold. This hierarchical processing allows the subsequent rasterization process to employ differentiated strategies for different data levels.

[0033] Dual-resolution grid rasterization: For the salient data point set, a first regular grid is constructed and rasterized using a higher spatial resolution; for the background data point set, a second regular grid with a lower resolution is constructed. Each data point is mapped to its corresponding grid cell, and the first and second regular grids are processed separately.

[0034] Grid cell aggregation and output merging: Within each grid cell, a maximum value aggregation strategy is applied to the data points mapped to that cell, selecting the value with the highest correlation significance as the representative value of that grid cell, and determining the corresponding color attribute. Finally, the rasterization results of the significant data point set are merged with the rasterization results of the background data point set to generate the final output result for visualization.

[0035] By following the steps above, background data can be compressed efficiently while preserving significant correlation peaks.

[0036] This method is applicable to most cases, focusing only on salient sites and suitable for rapid drawing applications where non-salient sites (background areas) are not critical. Examples of such applications are as follows:

[0037] Visualization of the genome-wide association analysis (GWAS) results based on duck plasma cholinesterase abundance (containing 12,232,312 marker sites) typically takes 285.7 seconds using conventional R language and the ggplot2 package. However, the visualization results obtained using this invention's multi-level rasterization method are shown below. Figure 2 The total calculation time was only 7.6 seconds, while ensuring high-quality display of significant sites.

[0038] Example 2: A Smart Rasterization Method Based on a Single Mesh

[0039] This embodiment provides an intelligent rasterization method based on a single regular grid, suitable for association analysis data with multiple chromosomes or large data scale. See the detailed flowchart below. Figure 3 .

[0040] Regular grid construction and data mapping: Construct a single regular grid within the target display space and map all input data points to the corresponding grid cells.

[0041] Grid cell feature calculation: For each grid cell, calculate the feature parameters: the number of data points in the grid cell and the maximum value of the correlation significance within the grid cell.

[0042] Intelligent output determination: According to preset rules, the following determination is performed on each grid cell: when the maximum correlation significance value in the grid cell exceeds the significance threshold, the grid cell is marked as output state; when the number of data points in the grid cell exceeds the density threshold, the grid cell is marked as output state; otherwise, the grid cell is randomly sampled to determine whether to output.

[0043] Representative value generation: For grid cells marked as output states, a maximum value aggregation strategy is used to generate representative values, and the corresponding visualization data is output. This embodiment effectively reduces memory usage while ensuring the complete preservation of significant signals, making it suitable for multi-chromosome parallel rendering scenarios.

[0044] This method is suitable for applications where there are high requirements for non-significant sites (background regions). Examples of such applications are as follows:

[0045] Visualization analysis of the genome-wide association study (GWAS) results of betaine abundance in beef muscle (containing 8,221,245 marker sites) typically takes 234.5 seconds using R language and the ggplot2 package. However, the visualization results obtained using the intelligent rasterization method described in this invention are shown below. Figure 4The total computation time is only 9.6 seconds, while ensuring high-quality display of significant sites. The advantage of this method compared to multi-level rasterization is that it meticulously maps even high-density regions containing sites with low P-values; therefore, the computation time will be slightly longer.

[0046] Example 3: Adaptive Density Rasterization Method

[0047] This embodiment provides an adaptive rasterization method based on data density. A detailed flowchart can be found below. Figure 2 .

[0048] Basic density statistics: First, a basic regular grid is constructed, and the number of data points in each basic grid cell is counted to generate a density distribution map.

[0049] Adaptive mesh subdivision: When the number of data points within a basic mesh cell exceeds a preset density threshold, the density threshold is calculated as follows:

[0050]

[0051] in This is an adjustment factor (default is 3). It subdivides the grid cell into multiple sub-grid cells; when the density does not exceed the threshold, the original grid structure is maintained.

[0052] Data redistribution and aggregation: Data points are redistributed to corresponding base grid or sub-grid cells, and a maximum value aggregation strategy is applied to the data points within each grid cell to generate the output result. This embodiment allows computational resources to be concentrated in high-density areas, thereby more realistically reflecting the data distribution characteristics.

[0053] This method is suitable for application scenarios that demonstrate the results of association analysis / molecular QTL combining multiple traits. Examples of such scenarios are as follows:

[0054] Based on the determination and analysis of genotype and cecal tissue gene expression in 515 ducks, eQTL mapping identified 14,022 genes in the duck cecal tissue that were cis-regulated by 8,920,919 markers, forming 31,375,588 marker-gene cis-regulatory pairs. Conventional R language + ggplot2 package plotting of these results requires 55.2 points, while the results obtained using the adaptive density rasterization method of this invention are shown below. Figure 6 The total calculation time is only 20 seconds, and high-quality display of significant sites is guaranteed.

[0055] Example 4: A QTL rasterization method for molecular phenotypes with continuous color preservation

[0056] This embodiment provides a rasterization method that preserves continuous colors for molecular phenotypic QTL data containing continuous effect values.

[0057] Color interval discretization: The continuous effect value is mapped to several discrete color intervals (e.g., 15 intervals) according to its size, and each color interval corresponds to an effect value range.

[0058] Color-separated rasterization: For each color zone, rasterization is performed on the data points, and the maximum value aggregation strategy is used to generate intermediate results for the data points in the grid cells.

[0059] Result merging and recoloring: The rasterization results of each color range are merged, and the output data is reassigned with color information according to the original color range, thereby achieving high-fidelity preservation of continuous color gradients.

[0060] This method is suitable for applications that require displaying correlation analysis results of continuous color changes or two-dimensional QTL results of trans molecules. Examples of such applications are as follows:

[0061] Based on the determination and analysis of genotype and cecal tissue gene expression in 515 ducks, trans-eQTL mapping was used to identify 12,718 genes in the duck cecal tissue that were cis-regulated by 1,042,153 markers, forming 2,346,170 marker-gene trans-regulatory pairs. Conventional R language + ggplot2 package plotting of the results takes 183.3 seconds, while the results of this invention using a continuous color-preserving molecular phenotypic QTL rasterization method are shown below. Figure 8 The total calculation time is only 5 seconds, and the high-fidelity preservation of the continuous color gradient of the site is guaranteed.

[0062] As can be seen from the above embodiments, the present invention can adaptively select rasterization and aggregation strategies according to different data scales, data densities and data types. While ensuring the complete preservation of significant association signals, it can greatly improve plotting efficiency and reduce memory consumption, and is particularly suitable for large-scale genome association analysis and visualization of molecular phenotypic QTL data.

Claims

1. A method for intelligent adaptive rasterization of genome association analysis and visualization of molecular phenotype QTL data, characterized in that, Includes the following steps: S1. Obtain association data generated by genome association analysis or molecular phenotypic quantitative trait locus analysis, wherein the association data includes at least genomic location coordinates and association significance indicators; S2. Construct a regular grid for data mapping within the target display space, and map the associated data to the corresponding grid cells; S3. Based on the numerical and spatial distribution characteristics of the associated data, the data within the grid cell is aggregated to generate at least one representative value; S4. Based on the representative value, determine whether to output the corresponding grid cell and generate output data for visualization. The aggregation process employs at least a maximum value aggregation strategy to preserve significant correlation signals. In step S3, the aggregation process includes multi-level rasterization processing, specifically: the associated data is divided into a set of salient data points and a set of background data points according to a preset saliency threshold, and rasterization processing is performed using grids of different resolutions respectively. In step S4, whether to output a grid cell is determined based on at least one or a combination of the following rules: when the maximum correlation significance value within the grid cell exceeds a significance threshold, the grid cell is output; when the number of data points within the grid cell exceeds a density threshold, the grid cell is output; when the number of data points within the grid cell does not exceed the density threshold, the grid cell is randomly sampled and output. After constructing the regular grid, the data point density within each grid cell is calculated. When the data point density of a grid cell exceeds a preset threshold, the grid cell is subdivided to form multiple sub-grid cells.

2. The intelligent adaptive rasterization method for genome association analysis and molecular phenotype QTL data visualization according to claim 1, characterized in that, The salient data point set is rasterized using a first grid resolution, while the background data point set is rasterized using a second grid resolution lower than the first grid resolution.

3. The intelligent adaptive rasterization method for genome association analysis and molecular phenotype QTL data visualization according to any one of claims 1 to 2, characterized in that, For each data point within a grid cell, the maximum correlation significance value is used as the representative value, instead of calculating the average value of the data points.

4. The intelligent adaptive rasterization method for genome association analysis and molecular phenotype QTL data visualization according to claim 1, characterized in that, Data points within the subdivided subgrid cells are aggregated separately, while the unsubdivided grid cells retain their original aggregation method.

5. A visualization system applied to the intelligent adaptive rasterization genome association analysis and molecular phenotype QTL data visualization method described in claim 1, characterized in that, include: The data acquisition module is used to acquire correlation analysis data; The grid building module is used to construct regular grids and map data to grid cells; The aggregation processing module is used to aggregate data within the grid cells and generate representative values; An output control module is used to determine whether to output the corresponding grid cell and generate a visualization result based on the representative value; wherein, the aggregation processing module adopts at least a maximum value aggregation strategy.

Images