Scalable detection of decentralized genetic biomarkers in human microbial data to personalize medical treatment

Abstract

Detecting biomarkers in human microbiome DNA to predict a biological condition and administer its treatment. A microbiome network may be generated comprising nodes representing microbiome DNA sequences and edges representing a co-occurrence of each pair of microbiome DNA sequences in a same DNA sample or sub-length. Nodes may be bundled into distinct groups based on the node's degree quantifying a number of its edges indicating a number of unique microbiome DNA sequences that co¬ occur with the microbiome DNA sequence represented by the node in the same DNA sample or sub-length. Groups of bundled microbiome DNA sequences may be validated having an internal connectivity that satisfies an anomaly condition indicating the group's sequences co-occur with a probability that is unlikely randomly statistical, e.g., deviating from a power law distribution. A machine learning model may be trained with the validated groups to predict a biological condition correlated therewith to administer its treatment.

Description

P-655574-PCSCALABLE DETECTION OF DECENTRALIZED GENETIC BIOMARKERS IN HUMAN MICROBIAL DATA TO PERSONALIZE MEDICAL TREATMENTFIELD OF THE INVENTION

[0001] Embodiments of the present invention relate generally to the field of personalized medical treatment based on microbiome DNA to improve treatment efficacy. Some embodiments of the invention relate to analyzing microbiome DNA to predict biological conditions (e.g., treatment-response, medicinal efficacy, disease risk) in human hosts. In particular, embodiments of the present invention relate to the efficient detection of decentralized genetic biomarkers in human microbial DNA. Embodiments of the invention provide a scalable, unsupervised method for identifying genetic information based on groups of k-mers that exhibit statistical significance in co-occurrence, enabling biomarker discovery at scale. Detection of those biologically-correlated biomarkers in patients enables administering treatments for those biological conditions to patients with improved treatment efficacy.BACKGROUND OF THE INVENTION

[0002] The human microbiome plays a fundamental role in various physiological processes, including digestion, immune response, and even neurological function. Microbiome comprises genetic material that may be obtained from microbial biological samples from various body sites, such as the gut (feces), skin, saliva, vaginal microbiome, and other sources. Microbiome genetic material contains nucleic acid sequences that may detected by sequencing (e.g., metagenomic sequencing also known as "shotgun sequencing" or metatranscriptomic sequencing) to analyze the microbial DNA. However, sequencing disrupts the natural order of genetic material, producing unordered short reads, which challenges traditional analytical methods.

[0003] Existing microbiome analysis techniques often rely on reference databases of known microbes as a way to restructure fragmented sequencing data. By mapping short reads to known microbial genomes, researchers can leverage algorithms that operate on long sequences of data, such as those used in large language models (LLMs). This approach enables structured interpretation but is inherently limiting because it restricts all discoveries to a small fraction of already characterized microbes. Given that millions of microbial species remain uncharacterized, this reference-based approach fails to capture the vast diversity present in microbiomes.P-655574-PC

[0004] A second common strategy in microbiome research is focusing on known genes. Researchers often investigate genes with established biological functions, allowing them to infer microbiome activity based on domain knowledge. This approach is beneficial because it provides immediate interpretability and actionable insights. However, it also constrains discovery - scientists cannot simply analyze microbiome data without prior hypotheses. The reliance on known genes means that the field is largely driven by predefined expectations rather than open-ended exploration, making it difficult to identify novel genetic interactions and functions.

[0005] An alternative, data-driven bioinformatics approach seeks to identify patterns directly from sequencing data without prior biological assumptions. This method scans sequencing datasets for statistical features that correlate with phenotypic traits or environmental conditions. However, due to computational constraints, this approach is currently limited to short k-mers (≤7 nucleotides). The combinatorial explosion of possible k-mer sequences makes it infeasible to analyze longer patterns, thereby limiting the depth of insights that can be derived.

[0006] In particular, human microbiome DNA samples are sequenced into a plurality (e.g., tens to millions) of “reads,” each representing a continuous sequence of (e.g., fixed or variable length, such as, 100-150) nucleotides. Each read is then sub-divided into a plurality of “k-mers,” each representing relatively shorter continuous fixed or variable length sequences (e.g., a fixed-length integer number k, such as, 30; or variable length including k=30, 60, 72, etc.) of the read’s nucleotides. For example, each DNA sequence position can be one of four nucleotides (A, T, C, or G), so the total number of possible k-mers of length k in the microbiome DNA is 4k. Experiments indicate k-mer lengths of 30-60 nucleotides associate with optimal phenotypic expression of biomarkers (e.g., shorter k-mer lengths often suffer higher false positives due to a higher likelihood of randomly appearing and longer k-mer lengths often suffer higher false negatives as longer sequences obfuscate or dilute significant segments). With 430or more possible 30-mer combinations per DNA sample, there are too many k-mer combinations to practically model correlations between k-mers and biological effect in the human host. Groups of multiple k-mers, the combination of which often correlates with biological expression, compounds this problem as there are exponentially more combinations of k-mer groups than individual k-mers (e.g., for k-mers of length k, there are 22*kpossible k-mers, approximately (22*k)2possible pairs, and 2A(22*k) possible groups of k-mers, which for k-mers of length k=30, would be 260= approximately 1018possible k-mers and 21000000000000000000possible k-mer groups). Such massive numbers of combinations of groups of k-mers makes it realistically impossible to model their correlation to biological effect.P-655574-PC

[0007] Accordingly, there is longstanding need inherent in the art for, and a wealth of knowledge to be gained from, efficiently modelling the biological effect of groups of k-mer or other nucleotide sequences in human microbiome DNA.SUMMARY OF THE INVENTION

[0008] Embodiments of the invention overcome this longstanding need inherent in the art by efficiently modeling correlations between human biological conditions and groups of k-mer or other nucleotide sequences in human microbiome DNA. A patient’s microbiome’s DNA containing those biological condition-correlated k-mer groups may thus prompt administering a treatment to improve that biological condition for improved personalized medical treatment.

[0009] Embodiments of the invention provide a scalable, unsupervised system and method for detecting biologically relevant groups of k-mers without requiring reference genomes or predefined gene functions. Unlike conventional approaches that are limited to short k-mers, embodiments of the invention may analyze long k-mers, vastly improving the descriptive capability of detected biomarkers. By freeing analysis from the constraints of known microbes, embodiments of the invention can automatically detect biomarkers even in previously uncharacterized microbial species. Moreover, it is uniquely capable of identifying decentralized biomarkers - functional units that are scattered across multiple microbes -enabling the discovery of complex genetic interactions that traditional approaches would miss. This cross-context applicability arises because genetic and epigenetic regulatory elements often span multiple microbial ecosystems within the human body.

[0010] Embodiments of the invention leverage empirical observations revealing the discovery that the degrees of nodes in a network created by connecting co-occurring k-mers follow a power-law distribution in human genome samples (see e.g., human DNA data in Figs. 4-6). This fundamental property was not previously known and provides a strong theoretical foundation for the clustering and validation process used to detect the biomarkers.

[0011] A system, device, method and non-transitory computer-readable storage medium comprising instructions that when executed cause one or more processors to accurately and efficiently detect biomarkers that are groups of DNA sequences in microbiome DNA of one or more human hosts to predict a biological condition. A plurality of microbiome DNA sequences (e.g., DNA reads or k-mers) may be stored that are (e.g., metagenomically or metatranscriptomic) sequenced from multiple microbial organisms in microbiome DNA of one or more human hosts. A microbiome network may be generatedP-655574-PCcomprising a plurality of nodes representing the respective plurality of microbiome DNA sequences and a plurality of edges representing a co-occurrence of each pair of microbiome DNA sequences in a same sample or sub-length of the human microbiome DNA (e.g., reads in string or k-mers in reads). Microbiome DNA sequences may be bundled, into distinct (e.g., disjoint or overlapping) groups (e.g., candidate or potential anomalous groups), that are represented by nodes based on the node’s degree, the degree of each node quantifying a number of edges that contain the node indicating a number of unique microbiome DNA sequences that co-occur with the microbiome DNA sequence represented by the node in the same sample or sub-length of the microbiome DNA. Groups of microbiome DNA sequences may be validated as anomalous that each have an internal connectivity between nodes of the same group that satisfies an anomaly condition indicating the microbiome DNA sequences of that group co-occur with a probability that is unlikely randomly statistical. The internal connectivity condition is satisfied if the internal connectivity for a group of nodes deviates from an expected internal connectivity in a network that follows a power law distribution. A training dataset may be generated correlating the validated anomalous groups of microbiome DNA sequences from one or more human hosts with a biological condition measured in one or more same or different human hosts. A training phase may be executed in which a machine learning model may be trained with the training dataset to input microbiome DNA sequences from one or more human hosts and predict the biological condition for one or more same or different human hosts. A predictive run-time phase may then be executed for a human patient in which the model inputs new microbiome DNA sequences from one or more human patients and outputs a prediction of the biological condition for the human patient if the patient’s microbiome DNA sequences comprise one or more of the validated groups of microbiome DNA sequences correlated with the biological condition measured in the one or more human hosts. If so, a treatment for the biological condition may be administered to the human patient predicted to have the biological condition to improve the biological condition and / or treat the biological condition with improved efficacy e.g., compared to average populations, control groups, and / or to patients without those biomarkers (or with fewer or different combinations of their validated groups).BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.P-655574-PC

[0013] The subject matter regarded as the invention is particularly pointed out and distinctly claimed in the concluding portion of the specification. The invention, however, both as to organization and method of operation, together with objects, features, and advantages thereof, may best be understood by reference to the following detailed description when read with the accompanying drawings in which:

[0014] Fig. 1 schematically illustrates a system for detecting biomarkers that are groups of DNA sequences in microbiome DNA of one or more hosts to predict a biological condition, according to an embodiment of the invention;

[0015] Fig. 2 schematically illustrates data structures representing a portion of a microbiome network of microbiome DNA sequences, according to an embodiment of the invention;

[0016] Fig. 3 is a graph of an adjacency matrix of a microbiome network clustered into groups, according to an embodiment of the invention;

[0017] Fig. 4 includes two graphs representing distributions of repetitions of reads in a DNA sequence combining sequences sequenced from multiple biological samples from multiple hosts, according to an embodiment of the invention;

[0018] Figs. 5-6 is two graphs representing distributions of repetitions of reads sequenced from the DNA of a single human host obtained from a distinct biological sample, according to an embodiment of the invention;

[0019] Fig. 7 is a graph of an analysis of a clustered group of microbiome DNA sequences labeled by circle “A” in Fig. 3, according to an embodiment of the invention;

[0020] Fig. 8 is a graph of an analysis of a clustered group of microbiome DNA sequences labeled by circle “B” in Fig. 3, according to an embodiment of the invention;

[0021] Fig. 9 is a flowchart of a method for detecting biomarkers that are groups of microbiome DNA sequences of one or more hosts to predict a biological condition, according to an embodiment of the invention;

[0022] Fig. 10 is a graph of the sampling depth of sequenced samples across various sequencing processes, according to some embodiments of the invention; and

[0023] Fig. 11 is an analysis of sequence duplication levels, illustrating the extent of duplication across various sequencing processes, according to some embodiments of the invention.

[0024] It will be appreciated that for simplicity and clarity of illustration, elements shown in the figures have not necessarily been drawn to scale. For example, the dimensions of some of the elementsP-655574-PCmay be exaggerated relative to other elements for clarity. Further, where considered appropriate, reference numerals may be repeated among the figures to indicate corresponding or analogous elements.DETAILED DESCRIPTION OF EMBODIMENTS OF THE INVENTION

[0025] Embodiments of the invention provide a new approach for generating a machine learning model trained to analyze microbiome DNA sequences of biological samples of human host(s) and predict correlations with biological conditions (e.g., the efficacy of drugs, disease risk, etc.) in host(s) that are trained based on the biological condition(s) measured in the host(s). In general, the predictive model may correlate microbiome DNA groups with, and predict, any measurable biological condition(s) of the host(s), such as, medicinal efficacy, gastrointestinal or overall health (e.g., microbe diversity), disease susceptibility (e.g., to diabetes), disease tolerance, likelihood of disease recovery (e.g., from COVID), life expectancy, and / or fatality risk. Host(s) may include an individual host or a group of multiple hosts (e.g., a genetic family, a demographic group (e.g., having a common age range, gender, etc.), sharing health risks or having similar genetic profiles). Predicting conditions for a group of multiple hosts may be achieved by individually predicting conditions for each of the multiple hosts and averaging hostspecific predictions (e.g., merging model outputs) or combining the DNA sequences of the multiple hosts in a “DNA soup” containing a combination of multiple or all hosts’ samples (e.g., merging model inputs). Additionally or alternatively, each host’s DNA sample may be represented by a network of nodes and DNA sequences from multiple hosts may be merged by super-positioning a plurality of networks each representing a distinct microbiome DNA sample to generate a composite network representing a plurality of microbiome DNA samples. Hosts sampled to generate training data inputs in the model’s training phase may be the same or different than hosts for which biological conditions are predicted as outputs in the model’s prediction phase. In the prediction phase, hosts sampled for the model’s input data may be the same or different than hosts for which biological conditions are predicted as the model’s output data. For example, hosts for which biological conditions are predicted may be in the same family or have a similar genetic profile or sequences as the sampled hosts. Different hosts may be distinct (disjoint or overlapping) groups of one or more hosts.

[0026] Due to the unmanageable number of different possible k-mers in microbiome DNA (e.g., 43030-mer combinations) and even larger number of groups of those k-mers (e.g., 24^3030-mer pair combinations), conventional models cannot analyze these volumes of genetic material to predict their biological effect in practical time. Analyzing an exponential number of combinations of groups is an NP-P-655574-PChard (exponential) problem (e.g., similar to the traveling salesman problem) that is impractical to solve in finite time. Embodiments of the invention solve this problem by providing an efficient two-stage technique to predictively model groups of multiple k-mers to correlate, in combination, to biological condition(s) of microbiome hosts. In a first stage of unsupervised learning, microbiome DNA sequences are analyzed to efficiently validate significant, anomalous or atypically appearing patterns of groups of DNA sequences (e.g., k-mers and / or reads). In a second stage of supervised learning, the manageable subset of validated anomalous k-mer groups are labeled with biological condition(s) measured in hosts having those k-mer groups, thereby generating training data to train a model to predict the biological condition correlated with those anomalous groups of microbiome DNA sequences.

[0027] Step 1 identifies patterns of groups as anomalous or atypical based on a discovery that cooccurrences of k-mers in groups of microbiome DNA sequences follows a power law distribution. Embodiments of the invention may validate a subset of “significant” groups of microbiome DNA sequence that exhibit anomalous co-occurrence patterns that deviate from this power law distribution and so, their expression is unlikely randomly statistical, and therefore likely associated with a phenotypic property associated with a biological condition. The first stage thus eliminates non-validated non-anomalous groups of microbiome DNA sequences, which are likely randomly statistical noise. The training dataset in stage 2 thereby only includes the reduced manageable set of validated anomalous groups of microbiome DNA sequences, making it possible to train the machine learning model based on groups of microbiome DNA sequences in finite practical time. This first stage also reduces the size of the search space from previously unmanageable volumes of all combinations of microbiome DNA sequences groups (e.g., 24^3030-mer pairs) to a manageable volume of only bundled groups of validated anomalous microbiome DNA sequences (e.g., an order of magnitude of k-mer groups equal to the number of distinct degrees in a network, which depends on how the groups are bundled, and is less than a linear function of the number of individual k-mers in the network). A reduction in training data size from an exponential to a linear function of the number of individual k-mers in the network in stage 1 reduces the storage size and increases training speed by orders of magnitude, thereby improving training efficiency in stage 2.

[0028] The first stage may select the anomalous subset of the plurality of microbiome DNA groups as follows. A plurality of microbiome DNA sequences, such as reads and / or k-mers, may be received and stored. In some embodiments, a pre-filtering process may reduce the total number of microbiome DNA sequences to include only repeat sequences (that occur greater than any predefinedP-655574-PCinteger threshold number of times, such as at least twice, in the microbiome DNA) and exclude nonrepeat or low-repetition sequences (that occur less than any predefined integer threshold number of times, such as only once or never, in the microbiome DNA). In one example, the filter may reduce the number of k-mers in the model from 430total possible 30-mers to an order of one million repeating k-mers. However, even after filtering, the number of groups of microbiome DNA sequences (e.g., > 2millionk-mer pairs) is too large to train a model in practical time.

[0029] Embodiments of the invention solve this problem by reducing the number of the sequence groups into a computationally manageable number of anomalous groups of microbiome DNA sequences or nodes. This reduction is performed by creating a microbiome network of DNA sequence nodes, then bundling network nodes into potentially or candidate anomalous groups, and then validating potentially anomalous node groups as validated anomalous node groups.

[0030] A microbiome network GX= G(V, EX) (e.g., a subregion of network 100 shown in Fig.2) may be generated comprising a set of a plurality of nodes V={vi} (e.g., nodes 101 of Fig. 2) representing the respective plurality of i unique microbiome DNA sequences (e.g., k-mers and / or reads) and a plurality of edges Ex={vi vj} (e.g., edges 103 of Fig. 2), each representing a co-occurrence of a unique pair vj and vj of microbiome DNA sequences in a same sub-length or sample of the microbiome DNA (e.g., two k-mers co-occurring in the same read and / or two reads co-occurring in the same DNA sample sequence). In some embodiments, sequences from multiple DNA samples may be combined by superpositioning a plurality of networks each representing a distinct microbiome DNA sample to generate a composite network representing a plurality of microbiome DNA samples (e.g., where different samples are collected from the or different genetic families, populations or groups of human hosts).

[0031] Bundling is performed by sorting nodes in the network into distinct (disjoint or overlapping) groups based on the degree of the group and nodes therein. A node’s degree may quantify a number of edges connecting that node (e.g., a number of other unique microbiome DNA sequences that co-occur with that node’s microbiome DNA sequence in the same read or DNA sample). Groups may bundle nodes that have the same and / or similar (e.g., + / - an integer degree), and the group’s degree may be a dynamic or tunable parameter. This bundles together a group of all sequences (or associated nodes) that co-occur in a DNA sample or sub-length with the same (and / or similar) number of unique other sequences (or associated nodes). A group may bundle DNA sequences that are decentralized (e.g., from different microbes) or centralized (e.g., from the same microbe) in the DNA sample to detect biological effects of decentralized or centralized biomarkers and, additionally or alternatively, may bundle DNA sequencesP-655574-PCthat are contiguous, non-contiguous or a combination of contiguous and non-contiguous in the DNA sample. Microbiome DNA sequence and / or their corresponding nodes may be ordered, e.g., in a sequence, from groups of bundled nodes that are relatively more densely connected (higher degree) to relatively more sparsely connected (lower degree).

[0032] Validation may then be performed by selecting groups of nodes that have sufficiently deviant (too high or too low) internal connectivity indicating microbiome DNA sequences represented by the nodes co-occur in the same read or DNA sample with sufficiently deviant (high or low) frequency to indicate anomalous expression (e.g., deviating from the expected power law distribution). The internal connectivity of each group of the same degree may be computed, for example, to quantify how interconnected the sequences or nodes are in the same group. In various embodiments, the internal connectivity may be determined based on a number of nodes in the same group, number of edges in the same group, a number of pre-defined partially or fully connected node clusters (e.g., triangles or polygons or straws or cliques of nodes) in the same group or deviation therefrom. In one embodiment, the internal connectivity may be measured based on a ratio between the number of “internal edges” (edges connecting two nodes of same or similar degree) and the total number of “overall edges” (internal and external edges) for that group of the same degree. In one example, a maximal internal connectivity (e.g., normalized to 1) may indicate the group is a “clique” in which all nodes of that degree are connected to each other internally within the same group and to no other nodes external to the group), and a minimal internal connectivity (e.g., normalized to 0) indicates no internal group connections. Within maximal and minimal limits, a relatively higher internal connectivity indicates a relatively greater interconnection compared to external connection. Fig. 2 visualized the internal connectivity of each of a plurality of groups as the density of nodes in the group’s square (internal edges) vs. the total density of nodes in the column and row containing the square (overall edges). For example, groups with relatively high internal connectivity are represented by a relatively highly dense square surrounded by its relatively low density column. A threshold for validating groups as having anomalous inter-connectivity may be based on a discovery that standard groups of sequences or nodes co-occur in DNA reads or samples according to a power law distribution, e.g.: P(degree(v) = d)~d-α, where degree(v) is the number of neighbors node v has in the network, d is a given group degree, and α is a power-law parameter of the network. That is, the standard probability that a sequence or node co-occurs with a number d of different unique other sequences or nodes in the same DNA read or sample diminishes proportionally to a power of that number of co-occurrences. Figs. 3-6 show this power law relationship. Embodiments of the invention exploit thisP-655574-PCdiscovery by validating significant or anomalous groups of sequences or nodes whose internal connectivity or sequence co-occurrence in DNA deviates from (e.g., is significantly greater or less than) an expected internal connectivity in a network that follows a power law distribution, suggesting an abnormal abundance or absence of those groups of sequences or nodes indicating they are anomalous and candidates for correlation with a measured biological condition of the microbiome hosts. In an embodiment (e.g., in theorem 5.8), the internal connectivity condition for a group of nodes having a samedegree d may be that the group’ s internal connectivity is greater than or equal to:’ *, where xthresis a threshold in which the probability of observing more than the threshold is less than a value e (e.g., a fixed or tunable confidence parameter) and | | is the number of nodes in the microbiome network of degree d.

[0033] In some embodiments, when degrees are too small or too large, groups tend to deviate from an expected neutral node co-occurrence that follows a power law distributions. Accordingly, an additional or alternative pre-validation stage may filter microbiome DNA sequences or groups to include only nodes within a predefined degree range (e.g., > dmin and / or < dmax, where dmin and dmax are predefined minimum and maximum threshold degrees such as an integer greater than 2, 3,..., m), and edges connecting those nodes, thereby eliminating nodes in the network with degree outside that range and the edges connected to those eliminated nodes. Excluding groups with extreme degrees may cause retention of (only) groups that follow the power law distribution to allow more accurate validation of anomalous groups.

[0034] After applying the threshold to select the validated anomalous groups of sequences or nodes (e.g., indicating the co-occurrence of the group of nodes is unlikely), these sequences or nodes are used to train the predictive model in the second stage.

[0035] In the first stage, bundling network groups reduces the number of microbiome DNA sequences groups to analyze from an unmanageable number of all combinations of microbiome DNA sequences groups (e.g., the order of magnitude of k-mer groups is an exponential of the number of individual k-mers, such as, 2Ngroups of N k-mer pairs; so even after filtering the k-mers from 4kpossible k-mers to approximately 1 million repeating k-mers, the number of groups is 21,000,000which is still unmanageable) to a manageable number of bundled groups of microbiome DNA sequences (e.g., an order of magnitude of k-mer groups equal to the number of distinct degrees in a network, e.g., which depending on how the groups are bundled is at most a linear or square function of the number ofP-655574-PCindividual k-mers in the network). Consider a microbe network that has a number of nodes |V| and a number of edges |E|. If every node is bundled into one group, based on its degree, then the number of groups will be the number of degrees. The maximal possible degree is |V| (for a node that is connected to every other node) and the minimal possible degree is 1, so the number of different degrees is at most | V| (and probably significantly less). Accordingly, the number of groups would be at most |V|. If every node is bundled into multiple fixed number of groups, based on its degree, then the number of groups is | V| times a fixed number, which is also on the order of the number of nodes O(| V|). All these groupings based on nodes are linear to the number of nodes in the network. If groups are bundled based on edges, similarly, the number of groups would be on the order of the number of edges O(|E| ). Because the microbe network has a maximal number of edges that connects every node to every other node, |E| = |V|*|V|, generating a maximal possible number of groups on the order of a square of the number of nodes O(|V|A2). The precise group reduction may depend, e.g., on the network architecture (e.g., fully or partially connected, layer structures, how edges are counted such as between all nodes or only adjacent or up to nearby layers, how groups are constructed as only the same or including similar numbers of edges, case-specific network data such as the number or arrangement of nodes, etc.). The precise group reduction may also be a tunable design parameter, such that, degrees are defined to sort nodes into a predefined fixed or dynamic tunable number of groups. Reducing the number of microbiome DNA sequences groups being analyzed from an exponential to (e.g., less than) a linear or square power of the number of individual microbiome DNA sequences converts a previously NP-hard problem solvable in exponential time to a linear problem solvable in linear time. This boon to efficiency makes the previously impractical task of analyzing groups of microbiome DNA sequences computationally practical, and so also makes practical its use for training a predictive model to correlate those groups with biological conditions in human hosts, thereby revealing a wealth of new knowledge.

[0036] The second supervised learning stage trains a predictive model using a training dataset comprising anomalous microbiome DNA sequence groups (inputs) (or frequency, internal connectivity or scores thereof) correlated to biological conditions in human hosts (outputs). The training dataset may comprise only those selected groups validated as anomalous microbiome DNA sequences (e.g., a minority, such as, a relatively small number of groups, such as hundreds of thousands or millions, compared to 21000000possible groups) and excluding the remaining (e.g., vast majority of) non-validated groups of microbiome DNA sequences that either failed validation or were not bundled as candidates for validation. Generating the training dataset with (only) validated anomalous groups of microbiome DNAP-655574-PCsequences (excluding non-validated non-anomalous groups of microbiome DNA sequences) also significantly reduces the storage size of the training dataset (e.g., by a reduction factor of #krners / 2#kmers, such as, 1 O6 / 21’000’000or more) compared to a training dataset containing all sequence groups and thereby increases training speed and efficiency (e.g., by the same reduction factor) based on that training dataset to terminate in practical time. The predictive model may be trained to correlate any or particular threshold(s) of occurrence frequencies of validated groups with positive, neutral and / or negative (absolute or relative) biological conditions and / or degrees of biological conditions (e.g., of N-levels or a continuous scale of severity) and / or therapeutic response and / or degrees of therapeutic response to treatment of the biological conditions (e.g., of M-levels or a continuous scale of therapeutic responsiveness). Any number of single or multiple validated groups, types / levels / categories / occurrence frequencies, degrees of biological conditions and / or therapeutic responses, side-effects, safety factors, treatment-administration factors (e.g., dosages, timings for administering, combinations of drugs, etc.), patient factors (e.g., demographics, diagnosis or disease(s)), iterations of analysis, etc., or combinations thereof may be used. The correlations between any of these parameters or numbers thereof may be one-to-one, one-to-many, many-to-one or many-to-many. In various embodiments, there may be an integer number of N levels of biological conditions and / or M levels of their therapeutic response to treatment (e.g., degree ranges, such as, chemical blood levels, and / or each with or without treatment side-effects), a vector of NxM, NxP or MxP and / or a matrix of NxMxP biological conditions and / or therapeutic responses (e.g., for multiple N levels of biological conditions, M therapeutic response and / or side-effects and / or P safety measures), Q different biomarkers of validated groups (e.g., individually or in combination) with a frequency of occurrence that is statistically significant (e.g., by any degree relative to the power-law distribution) in patients with biological conditions), Fm features characterizing different statistical e.g. co-occurrence patterns of validated groups, etc. The multiple N-levels of biological conditions and / or M-levels of therapeutic response may be relative response measures (e.g., high, medium, low, no physiological benefit or Gaussian or other distribution range percentage levels), or quantitative physiological ranges (e.g., ten absolute value ranges or percentage improvements in levels of biological sample including but not limited to blood, saliva or feces). Any combination or division of these characteristics may be used. The predictive model may be trained by any machine learning model including, but not limited to, neural networks or deep learning, linear regression, logistic regression, support vector machines, ensemble learning, decision forests, K-mean clustering, and / or K-nearest neighbor algorithms.P-655574-PC

[0037] In some embodiments, stages 1 and 2 of the training phase may be repeated to retrain the model based on new microbiome DNA (e.g., for a periodic collection, new biological samples, etc.). In some embodiments, stages 1 and 2 may or may not be repeated at the same rate. For example, stage 1 may only be executed once or periodically and not for each model retraining (e.g., when stage 1 is executed globally based on inter-group hosts), whereas stage 2 may be executed for each model retraining (e.g., executed locally based (only) on hosts in a specific group, population or genetic profile for which prediction is sought).

[0038] After the model is trained in training phase, a predictive run-time phase may be executed in which the model inputs new microbiome DNA sequences from one or more host patients and outputs a prediction of the biological condition for one or more same or different host patients. Steps may be taken to change or preemptively avoid or encourage the predicted biological condition for one or more same or different hosts than the hosts for which the biological condition is predicted. For example, a first group of hosts (e.g., to which medicine was administered) having measured biological conditions (e.g., positive health benefits or reduced health detriments) may be used to train the model; a second group of hosts (e.g., to which no medicine was administered) may be predicted to have those biological conditions (e.g., positive health benefits or reduced health detriments); which may trigger a practice to alter or maintain the biological condition of a third group (e.g., to administer that medicine to), where all three groups are distinct (disjoint or overlapping), but for example, are associated by being in the same group (e.g., age range, gender, health risk category, from the same family or ancestry or have similar genetic profiles).

[0039] A treatment for the predicted biological condition may be administered to those patients whose microbial DNA comprise the one or more validated groups correlated with the biological condition. Additionally or alternatively, based on patient-detected validated groups, embodiments of the invention may administer or predict therapeutic response to treatments for the biological condition and / or different treatment variants, such as response of indication Y to different drug x regimens (e.g., N dosages and / or schedules may be administered to N patient cohort groups predicted to have N respective therapeutic responses, indications and / or side-effects), administer or predict therapeutic response to treatments for the biological condition such as indication Y to different combinations of drug x with other drugs (e.g., including combinations with immunomodulators, anti-inflammatory agents, corticosteroids, antibiotics, probiotics or other genomic or microbiome-modulating agents, and biologies targeting different pathways), stratify patients as responders and non-responders before administering theP-655574-PCtreatment, predict optimized timing or sequence of treatment administration, predict therapeutic response to the treatments for the biological condition in specific patient subgroups (such as treatment-naive vs. treatment-experienced patients, patients with early vs. late-stage disease, patients with comorbidities or co-medications) to assess the biomarkers as a predictive or companion diagnostic in a clinical trial, predicting long-term health outcomes of administering the treatment (e.g., including relapse rates and remission duration) in biomarker-stratified cohorts (e.g., to support regulatory submissions or labeling for biomarker-enriched indications), predict and administer minimal effective dosage or maintenance dosage adjustments, predict therapeutic response to different treatment formulations, such as drug dosages and / or schedules and administer treatment formulations based thereon, e.g. to identify or reduce the likelihood of treatment-related adverse events or toxicity in biomarker-positive patients or reduce their exposure by not administering treatment to biomarker-negative patients, to guide retreatment, drug holiday, or step-down strategies based on biomarker-defined likelihood of sustained remission, to tailor the route of administration (e.g., oral vs. parenteral) based on biomarker status (e.g., especially if gut microbiome influences drug x metabolism or absorption), predict treatment resistance or secondary nonresponse emerging in specific biomarker-defined subgroups and administer treatments for the biological condition or not based thereon. Accordingly, based on the validated groups detected in biological sample(s) of host patients, embodiments of the invention may administer the treatment to the patients based on the biological condition(s) correlated with their validated groups. Thus, a patient with validated groups correlated with biological condition(s) may be administered a treatment for the biological condition(s) with improved efficacy, e.g., compared to average efficacy of random patients, control groups, and / or to patients without those biomarkers (or with fewer or different combinations of their validated groups). Conversely, a patient without those validated groups (or with treatment-averse validated groups) may be predicted to not have the biological condition(s) and may not be administered a treatment for the biological condition(s), may be administered an altered treatment (e.g., altered drug administration regimen, such as dosage, timing, drug combination, etc.), or may weigh against administering the treatment to the patient.

[0040] Embodiments of the invention may improve personalized medical treatment and patient safety by identifying and administering a treatment (e.g., drug x) selectively to patient(s) with validated groups trained to be correlated to a biological condition and / or a higher likelihood of treatment success and / or a lower likelihood, fewer or reduced levels or degrees of side-effects to the treatment for the biological condition (e.g., improving overall benefit considering side-effects), e.g., compared to averageP-655574-PCpopulations and / or patients without those validated groups. For example, embodiments of the invention may identify and administer a treatment for the biological condition selectively to patient(s) in which the likelihood of toxic or detrimental effects are outweighed by therapeutically or prophylactically beneficial effects of administering the treatment. Embodiments of the invention may (a) administer a treatment for the biological condition to a first set of patients (e.g., a testing or training population) (e.g., based on one or more dosages, schedules and / or combinations with other drugs); (b) determine validated groups of k-mers in treatment-responsive patients by analyzing whether the patients maintain or later develop (e.g., an improved or worsened) disease or indication; and (c) administer the treatment for the biological condition to a second set of patients (e.g., a treatment population) based on the presence or absence (e.g., or threshold amounts) of those treatment-linked validated groups (e.g., administered at treatment-linked or optimized dosages, schedules and / or combinations with other drugs). Because the second set of patients’ DNA has preferential treatment-responsive validated groups, their treatment may cause an improvement to their biological conditions compared to the first set of patients. Additionally or alternatively, embodiments of the invention may identify, select, or stratify patients for patient cohorts who qualify patients for treatment (e.g., with drug x), or who need modified treatment, or with a higher therapeutic index to the treatment. Embodiments of the invention may identify, select, or stratify patients for administering the treatment in clinical trials or other studies with the treatment, thereby increasing the likelihood that the clinical trial or study will succeed and / or for optimizing or improving the treatment regimen, e.g., to select a drug’s dosages, dosing frequency, route of administration, etc.

[0041] Reference is made to Fig. 1, which schematically illustrates a system for accurately and efficiently detecting biomarkers that are groups of DNA sequences in human microbiome DNA of one or more hosts to predict a biological condition, according to an embodiment of the invention.

[0042] The system of Fig. 1 may include a genetic sequencer 102 and / or a sequence analyzer 106. Units 102 and 106 may be implemented in one or more computerized devices as hardware and / or software units, for example, specifying instructions configured to be executed by one or more processors. One or more of units 102 and 106 may be implemented as separate devices or combined as an integrated device.

[0043] Genetic sequencer 102 may input DNA obtained from biological samples, such as, a microbiome sample of one or more real living human host organisms and may output each organism’s genetic sequence including the host’s genetic information at one or more genetic loci, for example, a plurality of microbiome DNA sequences. In some embodiments, genetic sequencer 102 may input aP-655574-PCbiological sample, perform shotgun fragmentation (e.g., randomly breaking up the genome into small DNA fragments), sequence each fragment individually, order the sequenced fragments by running a computer program to detect overlaps in the DNA sequences, and reassemble the fragments in their correct order to reconstitute the human DNA genome. Each single organism’s DNA sample may be sequenced for analysis as an individual set of microbiome DNA sequences or multiple organism’s DNA samples may be sequenced for analysis as a combined set of the host group’s microbiome DNA sequences.

[0044] Genetic sequencer 102 may store in one or more memorie(s) 114, and send to sequence analyzer 106 to receive and store in one or more memorie(s) 118, a plurality of microbiome DNA sequences (e.g., reads and / or k-mers), sequenced from microbiome DNA of one or more human hosts, for example, each microbiome DNA sequence having a fixed or variable length of nucleotides (e.g., 100-150 nucleotides in each read and / or k nucleotides in each k-mer, such as, k=30-60). To reduce the volume of the microbiome DNA sequences for analysis, e.g., prior to generating a microbiome network, genetic sequencer 102 and / or sequence analyzer 106 may filter the plurality of microbiome DNA sequences to include (only) microbiome DNA sequences that occur greater than a predefined integer number of times in the same sample or sub-length of the microbiome DNA, thereby including (only) sufficiently repeating microbiome DNA sequences and excluding non-repeating or insufficiently repeating microbiome DNA sequences.

[0045] Sequence analyzer 106 may generate a microbiome network (e.g., 100 of Fig. 2) comprising a plurality of nodes (e.g., 101 of Fig. 2) representing the respective plurality of microbiome DNA sequences and a plurality of edges (e.g., 103 of Fig. 2) connecting the nodes representing a cooccurrence of each pair of microbiome DNA sequences in a same sample or sub-length of the microbiome DNA (e.g., pairs of reads in a string of the entire or partial length of DNA or pairs of k-mers in the same read). Each host’s DNA sample may be represented by a network of sequence nodes, so combining multiple hosts DNA sequences may be performed by merging or super-positioning a plurality of host-specific to generate a composite network representing a plurality of microbiome DNA samples.

[0046] Sequence analyzer 106 may then bundle microbiome DNA sequences represented by nodes into distinct (e.g., disjoint or overlapping) candidate or potential anomalous groups, based on the node degrees. The degree of each node may quantify a number (e.g., or density, deviation from a mean or median thereof, etc.) of edges that contain the node indicating a number (e.g., or density, deviation from a mean or median thereof, etc.) of unique microbiome DNA sequences that co-occur with the microbiome DNA sequence represented by the node in the same sample or sub-length of the microbiomeP-655574-PCDNA. The degree condition for bunding nodes in the same group may be that the nodes have the same or similar degree, such as, + / - some integer or percentage deviation from the node’s degree. The degree condition for bunding nodes may be a tunable or dynamic parameter, e.g., manually set by a user or automatically set to optimize computer efficiency.

[0047] Sequence analyzer 106 may then validate candidate or potential anomalous groups into confirmed anomalous groups by measuring each group’s internal connectivity and selecting or validating groups of microbiome DNA sequences that each have an internal connectivity between nodes of the same group that satisfies an anomaly condition indicating the microbiome DNA sequences of that group cooccur with a probability that is unlikely randomly statistical. In some embodiments, the internal connectivity of each group may be determined based on a number of nodes in the same group, a number of edges in the same group, a number of pre-defined partially or fully connected clusters (e.g., triangles or polygons or straws (linearly connected) or cliques) in the same group or deviation thereof. Additionally or alternatively, the internal connectivity of each group may be determined based on a ratio between a number of internal edges connecting two nodes internal to the same group (e.g., visualized by the number or density of points inside each distinct box in the graph of Fig. 3) and a total number of overall edges connecting nodes within the group to any other node internal or external to the group (e.g., visualized by the number or density of points in rows and columns intersecting the group’s box in the graph of Fig. 3). In some embodiments, the anomaly condition is that the internal connectivity for the group of nodes deviates from an expected internal connectivity in a network that follows a power law distribution. Such a deviation from a power law deviation establishes a condition for the internal connectivity of the ratio above for a group of nodes having a same degree d, that the group’s internalconnectivity is greater than or equal to:, where xthresis a threshold in which the probability of observing more than the threshold is less than a value e (e.g., a fixed or tunable confidence parameter) and |V^|is the number of nodes in the microbiome network of degree d.

[0048] The combination of bunding and validating together provides an efficient and accurate technique for selecting anomalous groups of microbiome DNA sequences, and excluding non-anomalous groups of microbiome DNA sequences, to model those groups’ correlation with biological condition more accurately and efficiently. Without bundling, the number of combinations of groups to validate for anomaly conditions is on the order of an exponential of the number of groups (e.g., 2#nodes). Bundling nodes into disjoint groups reduces the number of combinations of groups to validate for anomalyP-655574-PCconditions to, for example, a linear order of the number of groups (e.g., O(#nodes)) or at most a square of the number of nodes (e.g., O(#nodes2), depending on the grouping). In some embodiments, more or less compact groupings may be used depending on network size, computer resources, convergence times, etc. For example, group numbers with a linear order of nodes may be used for larger networks, while group numbers with a square order of nodes may be used for relatively smaller networks. This reduction makes a realistically impossible NP-hard (exponential) problem (e.g., analyzing O(2#nodes) groups) to a realistically possible problem (e.g., analyzing O(#nodes) and / or < O(#nodes2) groups).

[0049] In some embodiment, groups with extreme degrees (greater than and / or less than predefined integer threshold degree(s), such as, outside 90% of the standard deviation from the mean degree) may deviate from an expected neutral node co-occurrence that follows a power law distributions used to validate groups to train the model. In one embodiment, prior to bundling and / or validation, sequence analyzer 106 may filter the plurality of microbiome DNA sequences to include (only) nodes with degrees within a predefined range and edges connecting those nodes, eliminating nodes deviating from the predefined degree range and edges connected to the eliminated nodes. In another embodiment, other neutral distributions and thus, conditions, may for the basis for validating groups of these extreme degrees.

[0050] After validation, sequence analyzer 106 may generate a training dataset correlating the validated groups of microbiome DNA sequences from one or more first hosts with a biological condition measured in one or more of same or different second hosts (e.g., different hosts having a similar genetic profile).

[0051] Sequence analyzer 106 may then train a machine learning model with the training dataset to input microbiome DNA sequences from one or more third hosts and predict the biological condition for one or more of same or different fourth hosts (e.g., different hosts having a similar genetic profile). Training may be repeated, e.g., periodically, based on new microbiome DNA.

[0052] Sequence analyzer 106 may then run a prediction phase in which the model inputs new microbiome DNA sequences from one or more fifth hosts and outputs a prediction of the biological condition for one or more same or different sixth hosts. Steps may be taken to change or preemptively avoid or encourage the predicted biological condition for one or more same or different hosts than the hosts for which the biological condition is predicted.

[0053] Genetic sequencer 102 and sequence analyzer 106 may include one or more controller(s) or processor(s) 108 and 112, respectively, configured for executing operations and one or more memoryP-655574-PCunit(s) 114 and 118, respectively, configured for storing data such as genetic information or sequences and / or instructions (e.g., software) executable by a processor, for example for carrying out methods as disclosed herein. Processor(s) 108 and 112 may include, for example, a central processing unit (CPU), a digital signal processor (DSP), a microprocessor, a controller, a chip, a microchip, an integrated circuit (IC), an application-specific integrated circuit (ASIC), or any other suitable multi-purpose or specific processor or controller. The ASIC is a physical circuit specially customized for a specific artificial intelligence application or task, such as, executing a neural network (in training and / or prediction phases). The ASIC may comprise a plurality of neurons organized in an array and connected via a plurality of synaptic circuits, wherein each neuron comprises a memory (e.g., register), a controller (e.g., microprocessor), and at least one input, and each synaptic circuit comprises a memory (e.g., register) for storing a synaptic weight connecting a pair of neurons. The ASIC provides superior computing capabilities and reduced electricity consumption when executing the specific artificial intelligence application or task, e.g., the neural network, compared to traditional general-purpose CPUs. Processor(s) 108 and 112 may individually or collectively be configured to carry out embodiments of a method according to the present invention by for example executing software or code. Memory unit(s) 114 and 118 may include, for example, a random access memory (RAM), a dynamic RAM (DRAM), a flash memory, a volatile memory, a non-volatile memory, a cache memory, a buffer, a short term memory unit, a long term memory unit, or other suitable memory units or storage units. Genetic sequencer 102 and sequence analyzer 106 may include one or more input / output devices, such as output display 120 (e.g., such as a monitor or screen) for displaying to users results provided by sequence analyzer 106 (e.g., visualizing Figs. 2-8) and an input device 122 (e.g., such as a mouse, keyboard or touchscreen) for example to control the operations of the system and / or provide user input or feedback, such as, selecting one or more hosts, selecting input genetic sequences, selecting one or more tuning parameters for model training, inputting new training data and retraining, selecting biological conditions (e.g., from a predefined set) to train or predict, selecting training accuracy, iterations or times, etc.

[0054] Reference is made to Fig. 2, which schematically illustrates data structures representing a portion of a microbiome network 100 of microbiome DNA sequences, according to an embodiment of the invention. Microbiome network 100 typically has too many nodes, for example, millions, to illustrate in its entirety. Data structures described herein may be stored in one or more memor(ies) (e.g., memory unit(s) 114 and / or 118 of Fig. 1), may be generated and controlled by one or more processor(s) (e.g., controller(s) or processor(s) 112 of a DNA sequence analyzer 106 of Fig. 1), and any visualizations, orP-655574-PCdata thereof may be displayed on one or more display(s) (e.g., output display 120 of Fig. 1 or any user display).

[0055] Microbe network 100 comprises a plurality of nodes 101 representing a respective plurality of microbiome DNA sequences, each sequenced from a biological sample of one or more human hosts. The plurality of nodes 101 may be pairwise connected by a plurality of edges 103. Each edge 103 may represent a co-occurrence of each pair of microbiome DNA sequences in a same sample or sub-length of the microbiome DNA, such as, two reads in the same partial or whole string of DNA or two k-mers in the same read. Nodes 101 may be densely-connected by edges 103 (e.g., > 80% of nodes connected by edges), sparsely-connected by edges 103 (e.g., < 20% of nodes connected by edges) or moderately-connected by edges 103 (e.g., 20-80% of nodes connected by edges). Nodes 101 and / or edges may be filtered, e.g., to remove edges having extreme degree (e.g., > dmin and / or < dmax, where dmin and dmax are predefined minimum and maximum threshold degrees).

[0056] Nodes 101 may each have a degree counting the number of edges to which the node connects. Nodes 101 may be bundled into groups of the same or similar degree. Nodes 101 in the same group thus have the same or similar number of unique microbiome DNA sequences that co-occur with the microbiome DNA sequence represented by the node in the same sample or sub-length of the microbiome DNA. Each bundled group may be validated to determine if it has sufficiently anomalous behavior to correlate with host biological condition for training a machine learning model, for example, if the nodes 101 of the group are inter-connected with a probability that is unlikely randomly statistical (e.g., too high or too low, indicating the nodes’ co-occurrence is likewise unlikely randomly statistical, but significant, and therefore likely correlated with host biological condition). The anomaly condition may be, for example, that the internal connectivity for the group of nodes deviates from an expected internal connectivity in a network, such as, one that follows a power law distribution. Internal connectivity of nodes 101 by edges 103 may be a measure of a number of nodes 101 or edges 103 in the same group, a density of nodes 101 or edges 103 in the network 100 or region thereof, a number of pre-defined partially or fully connected node clusters (e.g., triangles or polygons or straws or cliques of nodes) in the same group, a ratio of intragroup edges (internal to a group) to inter-group edges or overall edges (external to a group or in the entire network or region thereof), or derivative or deviation therefrom. Groups of microbiome DNA sequences represented by nodes 101 invalidated anomalous groups may be used, together with biological conditions measured for their DNA’s hosts, to generate a training dataset to train a machine learning model to corelate groups of microbiome DNA sequences with their host(s)’ biological condition(s).P-655574-PC

[0057] Reference is made to Fig. 3, which is a graph of an adjacency matrix of a network of microbiome DNA sequences clustered into groups, according to an embodiment of the invention. Each point in the groups represents a node. Fig. 3 may visualize the network of microbiome DNA sequences represented by an adjacency matrix. The adjacency matrix may be an N * N matrix, such that each coordinate (nl,n2) is binary (e.g., either 0 or 1), representing whether or not the graph contains an edge between node nl and n2. The order to the nodes of the graph may be unordered, so that, the nodes can be sorted in many different ways - each associated with a different adjacency matrix (e.g., differing in the ordering of the nodes, but having the same number of edges, degree distribution and so on, since all these matrices still represent the same graph). One example way to sort the nodes is: (1) sort the nodes by their degree, and then (2) group the nodes based on similar degree values. Fig. 3 shows that when nodes are grouped by degree, the internal connectivity of groups is easily visualized. For example, relatively “dense” group clusters emerge (e.g., containing groups with a relatively high number of internal edges), such as, the square labeled by circles “A” and “B” (analyzed in reference to Fig. 7 and 8), and relatively “sparse” group clusters emerge (e.g., containing groups with a relatively low number of internal edges), such as, the square labeled by circle “C” In one embodiment, Fig. 3 visualizes an internal connectivity of each group as a ratio between the number or density of points inside each distinct box (e.g., a number of internal edges connecting two nodes internal to the same group) and the number or density of points in the rows and columns intersecting the group’s box (e.g., a total number of overall edges connecting nodes within the group to any other node internal or external to the group).

[0058] Reference is made to Fig. 4, which includes two graphs representing distributions of repetitions of DNA reads combining sequences sequenced from multiple biological samples from multiple hosts, according to an embodiment of the invention. The left graph may represent the relationship between the number of reads (x-axis) and the number of times those reads repeat (y-axis) in the combined DNA sequences of the multiple hosts. The right graph may represent the relationship between the probability (y-axis) that the number of repetitions for a given node is greater than or equal to x (x-axis) in the combined DNA sequences of the multiple hosts. The probability in the right graph decreases monotonically (e.g., since every nodes that repeat X+l times also repeats, by definition, X times), and there is some value for which it receives 1 (e.g., the minimal number of repetitions for the nodes in the network). Both graphs follow a power law distribution. Both graphs represent microbiome DNA from the human microbiome project.P-655574-PC

[0059] Reference is made to Figs. 5-6, which includes two graphs representing distributions of repetitions of reads sequenced from the DNA of a single host obtained from a distinct biological sample, according to an embodiment of the invention. In the left graph, each curve represents, for the DNA of a single distinct host obtained from a distinct biological sample, the relationship between the number of reads (x-axis) and the number of times those reads repeat (y-axis). The right graph may represent the relationship between the probability (y-axis) that the number of repetitions for a given node is greater than or equal to x (x-axis).

[0060] Whereas Fig. 4 shows a power law distribution for a DNA sequence of microbiomes from multiple samples combined together, Figs. 5-6 shows approximately the same power-law distribution for (almost) every single host’s sample individually (individual curve in Fig. 6).

[0061] Reference is made to Fig. 7, which is a graph of an analysis of a clustered group of microbiome DNA sequences labeled by circle “A” in Fig. 3 (e.g., generated by bundling and validated with an anomaly criterion), according to an embodiment of the invention. Fig. 7 shows four elements:1. A list of k-mers in the clustered group labeled by circle “A” in Fig. 3.2. A distribution of the total appearance of these k-mers among the tested hosts (e.g., approximating a statistically neutral Gaussian distribution).3. A distribution of drug efficacy among the tested hosts (e.g., the biological condition the model is trained to predict).4. Correlation between (2) distribution of the total appearance of these k-mers among the tested hosts and (3) distribution of the drug efficacy among the tested hosts. In the example of the group labeled by circle “A,” there is no significant correlation.

[0062] Reference is made to Fig. 8, which is a graph of an analysis of a clustered group of microbiome DNA sequences labeled by circle “B” in Fig. 3 (e.g., generated by bundling and validated with an anomaly criterion), according to an embodiment of the invention. Fig. 8 shows four elements:1. A list of k-mers in the clustered group labeled by circle “B” in Fig. 3.2. A distribution of the total appearance of these k-mers among the tested hosts (e.g., deviating from a statistically neutral Gaussian distribution).3. A distribution of drug efficacy among the tested hosts (e.g., the biological condition the model is trained to predict).4. Correlation between (2) distribution of the total appearance of these k-mers among the tested hosts and (3) distribution of the drug efficacy among the tested hosts. In the example of the groupP-655574-PClabeled by circle “B,” there is a significant correlation between these k-mers and the biological condition (e.g., drug efficacy). This group can thus be used for prediction.

[0063] Reference is made to Fig. 9, which is a flowchart of a method for accurately and efficiently detecting biomarkers that are groups of DNA sequences in microbiome DNA of one or more hosts to predict a biological condition, according to an embodiment of the invention. Operations described herein may be executed by one or more processor(s) (e.g., controlled s) or processor(s) 112 of a DNA sequence analyzer 106 of Fig. 1), data or data structures described herein may be stored in one or more memor(ies) (e.g., memory unit(s) 118 of a DNA sequence analyzer 106 of Fig. 1), and any visualizations, or data may be displayed on one or more display(s) (e.g., output display 120 of Fig. 1 or any user display).

[0064] In operation 910, a process or processor may store a plurality of microbiome DNA sequences (e.g., DNA reads or k-mers) that are sequenced from microbiome DNA of one or more human hosts.

[0065] In operation 920, a process or processor may generate a microbiome network comprising a plurality of nodes representing the respective plurality of microbiome DNA sequences and a plurality of edges representing a co-occurrence of each pair of microbiome DNA sequences in a same sample or sublength of the microbiome DNA (e.g., reads in string or k-mers in reads).

[0066] In operation 930, a process or processor may bundle, into distinct (e.g., disjoint or overlapping) groups (e.g., candidate or potential anomalous groups), microbiome DNA sequences that are represented by nodes based on the node’s degree. The degree of each node may quantify a number of edges that contain the node indicating a number of unique microbiome DNA sequences that co-occur with the microbiome DNA sequence represented by the node in the same sample or sub-length of the microbiome DNA.

[0067] In operation 940, a process or processor may validate groups of microbiome DNA sequences as anomalous that each have an internal connectivity between nodes of the same group that satisfies an anomaly condition indicating the microbiome DNA sequences of that group co-occur with a probability that is unlikely randomly statistical. The process or processor may determine that the internal connectivity condition is satisfied if the internal connectivity for a group of nodes deviates from an expected internal connectivity in a network that follows a power law distribution.

[0068] In operation 950, a process or processor may generate a training dataset correlating the validated anomalous groups of microbiome DNA sequences from one or more human hosts with a biological condition measured in one or more same or different human hosts.P-655574-PC

[0069] In operation 960, a process or processor may execute a training phase to train a machine learning model with the training dataset to input microbiome DNA sequences from one or more human hosts and predict the biological condition for one or more same or different human hosts.

[0070] In operation 970, a process or processor may execute a predictive run-time phase, in which the model inputs new microbiome DNA sequences from one or more patient hosts and outputs a prediction of the biological condition for one or more same or different patient hosts. In one embodiment, the model may predict the biological condition for a patient whose plurality of microbiome DNA sequences comprises one or more of the validated groups of microbiome DNA sequences (e.g. in one or more occurrence frequencies or distribution patterns) correlated with the biological condition.

[0071] In operation 980, a treatment for the biological condition may be administered to the human patient predicted to have the biological condition for personalized DNA-based medical treatment. The treatment may be administered by a doctor or other medical personnel, the patient, or automatically by an automated medical delivery device triggered by a treatment program, the prediction of operation 770, and / or an intermediary person or system confirming the prediction results. Because the patient is predicted to have the biological condition responsive to the treatment, administering the treatment to the patient results in an improvement to the biological condition and / or treatment efficacy e.g., compared to average populations, control groups, and / or to patients without those biomarkers (or with fewer or different combinations of their validated groups).

[0072] Other or different operations or orders of operations may be used and some operations may be omitted repeated, e.g., operations 910-960 may be repeated (e.g., periodically, upon receiving new data, etc.) to retrain the model with new microbiome DNA sequences stored in operation 910 from a new host biological sample.

[0073] Some embodiments of the invention adopt an unbiased metagenomic approach to create a model that determines the most suitable drug or treatment customized for an individual or group of humans, allowing for precision application based on individual microbiome profiles. The technique disclosed herein acknowledges the significant variation and multitude of contributing factors that lead to the diverse responses observed among hosts. Such embodiments exploit the rich biological information stored in microbiome DNA, for example, transforming the microbiome of a select few hosts, allowing for the prediction of biological effect, to determine the most effective drugs or treatments tailored to a specific group or general population.P-655574-PC

[0074] Some embodiments use a two-stage trial design targeting the prediction of the biological efficacy, e.g., of drugs or other treatments, using human microbiome data. The first stage may engage an unsupervised machine learning process, trained on a diverse dataset, that for example includes microbiome samples from a wide spectrum of humans across various groups. The second stage may use a smaller subset of humans, for example, whose biological condition has been tested and documented periodically, to implement supervised learning. The second stage may construct a predictive model that associates microbiome profiles with the effectiveness of the drugs or other treatments.

[0075] Benefits include, not only in the use of the microbiome as a predictive tool, but also in the capacity to make sense of its complex raw data. The microbiome, rich in diversity and complexity, conventionally presents a significant analytical challenge that until now has hindered its utility in such applications. To tackle this, a data-driven approach may be used powered by state-of-the-art artificial intelligence technology, creating an intelligent model that acknowledges the extensive variation and plethora of factors contributing to diverse responses observed among hosts. By leveraging the power of the microbiome and artificial intelligence, embodiments of the invention provide an accurate and effective solution for predicting biological effects of medical treatments tailored to the individual host or its group.

[0076] This technique may be scalable, applicable to predict any measurable biological condition, and provides potential for continuous predictive improvement as more data accumulates.

[0077] Reference is made to Fig. 10, which is a graph of the sampling depth of sequenced samples across various sequencing processes, according to some embodiments of the invention.

[0078] Reference is made to Fig. 11, which is an analysis of sequence duplication levels, illustrating the extent of duplication across various sequencing processes, according to some embodiments of the invention. Fig. 11 shows a majority of samples exhibiting a low count of duplicates, signifying the high quality of the sequencing process.

[0079] From Microbial Data to Drug Efficacy Prediction

[0080] Some embodiments of the invention may operate for example according to the following descriptions and definitions related to data handling, validation, training and prediction.

[0081] INPUT: Input for embodiments of the invention may include e.g. the following:1. Metagenomic Sequencing Data: Microbiome samples may be collected and sent for deep shotgun metagenomic sequencing to generate data e.g. as follows:P-655574-PCo One or more microbiome samples sequenced using any sequencing technology (e.g., Illumina NovaSeq 6000 with paired-end 150bp reads).o The data may originate from a single biological sample or from a mixture of multiple biological samples sequenced together.o The microbiome can be derived from feces, skin, saliva, vaginal samples, or any other site within the human body.o Metatranscriptomic or other types of genetic sequencing may also be used.2. List of Predefined Canonic k-mers:o A predetermined set of k-mers selected from a previous analysis (given that there are 4Ak possible k-mers, preselecting a relevant subset is crucial to reduce computational complexity and ensure meaningful analysis).3. Distance Parameter (D):o Defines the maximum allowable distance between k-mers within a read to consider them as co-occurring. Default value may be the full length of the read and may be set to any length less than that length.4. Edge Weight Filter (W):o A threshold for filtering out edges with insufficient weight (e.g., weak co-occurrence signals). Default value is e.g. 1 indicating all observed edges are initially included. 5. k-mer Length (k):o Specifies the length of k-mers to be analyzed.Each tested drug or therapy may be administered to groups of training and validation humans, with an additional group of control humans that received no treatment. Biological conditions may be measured from each human periodically, for calculating the drug or therapy efficacy for each group.

[0082] OUTPUT: Output for embodiments of the invention may include e.g. the following:

[0083] An output according to embodiments of the invention may include a list of biomarkers, where each biomarker includes a group of two or more k-mers that co-occur with statistical significance, indicating potential functional or epigenetic relevance. Each biomarker may be statistically associated with a target biological condition with some weight. Additionally, the output may include a formula to combine these weights with the relative frequency of these sequences in a given microbiome sample, generating a final aggregated score, such as, between 0 (no efficacy) and 1 (maximal efficacy for thisP-655574-PCdrug). Once this biomarker output is available, it can be applied to future microbiome samples directly, eliminating the need to re-run the previous steps.

[0084] These biomarkers may be generic and not necessarily specific to the site from which they were collected. For example, a biomarker identified in fecal samples may later be relevant for understanding skin-related conditions. This cross-context applicability arises because genetic and epigenetic regulatory elements often span multiple microbial ecosystems within the human body.

[0085] PROCESS:

[0086] A Process for Unsupervised Detection of Microbial DNA Patterns (Stage 1): Raw data may be obtained from the sequencing strings of 100-150 nucleotides, e.g., without undertaking any identification of microbes or strains. Network-oriented DNA analytics according to embodiments of the invention may be used to analyze the data from all collected microbiome samples. This resulted in numerous DNA patterns, each analytically determined to be unlikely to appear spontaneously in random genetic microbial samplings, and therefore likely associated with a phenotypic property. Stage 1 may function as a potent dimensionality reduction process, capable of operating on billions of raw 100-150 base-long strings. It may extract a computationally manageable number of groups of “statistically meaningful” substrings, avoiding bias towards predefined feature spaces, data pre-processing, or the semantics of the problem at hand. Stage 1 may be updated as new data becomes available, leading to the identification of new DNA patterns that contribute to the system’s predictive capabilities for the same or new properties. The Process for Unsupervised Detection of Microbial DNA Patterns may proceed by executing the following one or more operations, e.g.:1. Initialize an Empty Network N:o Create a (e.g., n*n) sparse weighted adjacency matrix, where n is the number of canonic k-mers (e.g., n is an integer).o Sparse representation ensures minimal space requirements, leveraging the power-law degree distribution of biological sequence networks.2. Process Each Sample and Read:o For each read, extract all k-mers.o Ignore k-mers that are not in the predefined list of canonic k-mers.o For each pair of k-mers within distance D: connect the pair of k-mers with an edge in the network N.3. Apply Edge Weight Filtering:1P-655574-PCo Remove edges with weights below a predefined threshold W to eliminate noise.4. Group and Validate:o Use a clustering approach to identify groups of k-mers with statistically significant internal connectivity. A group may be considered a biomarker if its co-occurrence probability deviates from the expected power-law distribution of random sequences. o An example operation:1. Iterate over each degree in the network N, identifying nodes that share the same degree.2. For each degree group, compute the ratio of inward-pointing edges to outwardpointing edges for each node. Inward-pointing edges refer to edges connecting nodes within the same degree group, while outward-pointing edges refer to edges connecting nodes in the degree group to nodes outside of it.3. The degrees of nodes in the network follow a power- law distribution, meaning that the probability P(d) of a node having degree d follows the function:P(d) ∝ d^(-γ)where γ is a constant that characterizes the network structure.4. Using this property, the threshold may be defined for the inward-to-outward edge ratio as:θ(d) = E_in(d) / E_out(d)where E_in(d) is the number of inward-pointing edges for nodes of degree d and E_out(d) is the number of outward-pointing edges. The threshold is derived analytically based on the expected power- law distribution, for example, xthres.5. A group of nodes that exceeds this threshold is considered “validated” and classified as forming a valid cluster.6. Extract the clusters of k-mers that pass validation and designate them as biomarkers.This process serves a dual purpose: first, it is an extremely efficient mechanism for defining potential clusters, as identifying nodes for each degree is computationally simple and efficient; and second, it effectively filters out noise, ensuring that only groups of k-mers exhibiting statistically significant cooccurrence patterns are considered biomarkers, eliminating or reducing random associations.P-655574-PC

[0087] A Process for Filtering the Microbial DNA Patterns using Semantic Labels (Stage 2): For each DNA pattern (e.g., a collection of 100-150 long DNA bases), embodiments of the invention may filter only those whose frequency (e.g., the number of times they occur in a sample) correlates strongly with the biological condition to be predicted (e.g., a drug’s efficacy, defined by its change on biological condition, normalized for the control group). Stage 2 may be executed once for each group of labels (e.g., once per drug or treatment).

[0088] IMPROVEMENTS: Embodiments of the present invention may provide one or more of the following improvements to the state of the art:1. Descriptive Power of Biomarkers:• Traditional biomarker discovery methods rely on individual genes or short sequence motifs.Embodiments of the present invention identifies functional groups of k-mers, providing richer biological insights.• This improvement is further strengthened by the ability to incorporate long k-mers (e.g., large values of k, such as 30 or more).2. Unsupervised Learning at Scale:• Unlike conventional methods requiring labeled training data, embodiments of the present invention may be entirely unsupervised, enabling the discovery of biomarkers without predefined hypotheses. Additionally or alternatively some supervised learning may be used.3. Integration with Machine Learning Models:• The extracted biomarkers can be used as features in machine learning models. Examples of feature extraction include e.g.:o Counting the mean, median, or minimum occurrences of the biomarker’ s k-mers in a sample.o Constructing feature vectors based on the presence or absence of biomarkers across multiple samples.4. Scalability and Efficiency:• Embodiments of the present invention may process vast amounts of sequencing data efficiently by e.g.:o Operating on a sparse (e.g., nxn) network rather than an infeasible combinatorial space of k-mer pairs (e.g., where n is an integer). One or more processors (e.g., controller(s) or processor(s) 112 of a DNA sequence analyzer 106 of Fig. 1) may detect biomarkersP-655574-PCoperating on the sparse network in practical time (e.g., a few hours or days), compared to an impractical time (e.g., several days or more) to analyze the entire space of k-mer pairs.o Leveraging power-law distributions to reduce noise and enhance signal detection for more accurate biomarker detection.5. Decentralized Biomarker Discovery:• Traditional biomarkers are often associated with specific genes or organisms. Embodiments of the present invention may detect decentralized functional units - groups of k-mers that may be distributed across multiple microbes.• For example, a functional metabolic pathway could involve multiple microbial species, with relevant k-mers appearing in different genomes. Existing methods would miss such patterns, whereas embodiments of the present invention capture those patterns by analyzing cooccurrence relationships.6. Improved Personalized Medical Treatment• New biomarkers, e.g., detected due to reduced noise as well as complex and decentralized k- mer groups, improves genetic detection of biological conditions (e.g., and its treatment responsive or non-responsive variants) correlated with those biomarkers. Improved prediction of biological conditions and its therapeutic response to treatments improves the efficacy of administering treatments to patients based on their treatment-correlated biomarkers, thus improving personalized medical care.

[0089] CONCLUSION: Embodiments of the present invention introduce a powerful, scalable approach to microbiome biomarker discovery, overcoming the limitations of traditional reference-based and single-marker methods. By detecting functionally significant groups of k-mers, such embodiments enable a deeper understanding of microbial interactions to improve their impact on human health. The method's efficiency and unsupervised nature make it a valuable tool for large-scale microbiome research and precision medicine applications.

[0090] Microbial Data Analytics

[0091] Embodiments of the invention are predicated on the analysis of numerous microbiome samples collected from human subjects across diverse conditions and demographics (e.g., ranging from different geographical locations, ages, genders, lineages, and / or genetic profiles). In experiments, a subset of these subjects have been administered a drug or treatment, and subsequent biological condition valuesP-655574-PCwere measured, creating an experimental group, while others remained as a control group. Each sample encapsulates a plethora of ’’reads,” each representing sequences of e.g., 100 to 150 nucleotides. Embodiments of the invention aim to identify significant microbial genetic patterns pertinent to a biological condition or trait of interest, e.g., the high or low efficacy of a drug.

[0092] A “k-mer” is a contiguous sequence of k nucleotides derived from a larger DNA or RNA sequence, whether originating from microbial genomes, host genomes, or mixed samples. It serves as a fundamental unit for analyzing genetic material in bioinformatics. For example, given the DNA sequence " ATCGGAC", extracting 3-mers (k=3) results in " ATC", " TCG", " CGG", " GGA", and " GAC". The choice of k determines the granularity of sequence analysis, where larger k values capture more context but increase computational complexity.

[0093] In a formal mathematical description, an original longer sequence of nucleotides may be denoted as the string S and its length as n, and a k-mer may be a substring of S of length k. Given S[z j] that denotes the substring of S starting at position z and ending at position j (e.g., inclusive), a k-mer of S starting at position z may be denoted as. S[i : i + k-1]. Note that the starting position z may satisfy 1 ≤ i ≤ n-k+1 to ensure the substring of length k can be obtained from S. Consequently, the total number of distinct k-mers that can be extracted from a sequence S of length n is n - k + 1.

[0094] Furthermore, considering the biological context where each position in the string can be one of four nucleotides (e.g., A, T, C, or G in DNA or A, C, G, or U in RNA), the total number of possible k-mers of length k, without considering any specific longer sequence, is 4k.

[0095] Embodiments of the invention may be characterized by an unbiased exploration of large k-mers, e.g. those with k = 30, though not confined to this value. Previous studies have illustrated the optimal expressivity of k-mers of length 30 or longer for predictive applications. However, their use is typically constrained to cases of extreme data sampling or pre-set filtering criteria, both of which can introduce bias. Conversely, models that leverage k-mers as features in machine learning typically limit k to values of 6 or less, driven by concerns of data scarcity and potential model overfitting. Traditionally, an unbiased analysis of longer k-mers would be considered computationally impractical due to the vast number of possible combinations, approximately 260. Additionally, it would necessitate significant amounts of data to circumvent overfitting.

[0096] Embodiments of the invention leverage the understanding that the distribution of k-mers within DNA does not typically follow a uniform pattern but instead conforms to a power-law. This property allows embodiments of the invention to implement efficient analytic techniques and extract aP-655574-PCsignificant number of k-mer groups automatically. Each of these groups may be verified as likely associated with a particular biological condition or epigenetic trait. However, the relevancy of such traits to our current interest may vary.

[0097] Embodiments of the invention may manifest in a dual capacity. Firstly, some embodiments extend our analysis beyond merely long k-mers, thereby enhancing their expressivity, to encompass groups of k-mers, which, in turn, fortifies their role as potent predictive features. Secondly, some embodiments address data paucity by utilizing a technique that capitalizes on the power-law distributed data property, as opposed to a brute force examination of ’’all k-mers” or ’’all groups.”

[0098] This technique facilitates the efficient detection of ’’correlated anomalies” localized groups giving rise to network structures which do not naturally arise in power-law networks. Analytically, the presence of such groups is indicative of an underlying causality within the data, signifying an association with a specific property relevant to the group of genetic information.

[0099] Additionally, the nature of this approach, based on the holistic examination of microbiome samples, allows for each group of k-mers to potentially comprise DNA fragments derived from heterogeneous sources. This denotes that functionalities emanating from diverse microbes may concurrently contribute to the observed behavior of interest.[000100] Workflow[000101] The following workflow describes an embodiment of the invention for analyzing the microbial data:1. Represent each raw sequenced microbial data set as a network, e.g., with a fixed number of nodes M, but with varying configurations of edges.2. Create superposition networks by overlaying networks according to certain categories (e.g., samples from the same host group).3. Perform an unsupervised analysis of each of the original one-sample networks and the superposition networks. For each network, execute the following steps:a. For each unique degree d in the network, examine all nodes with degree d. b. Empirically calculate the “internal connectivity” of this group of nodes (e.g., the ratio between the number of edges starting and ending at nodes within this group and the number of edges starting in this group and ending outside of it).c. If this ratio exceeds an anomaly criterion (e.g., set by Theorem 5.8), analytically determine that this group of nodes (e.g., k-mers) is likely to be statisticallyP-655574-PCsignificant with some statistical confidence e (e.g., a selected and / or tunable parameter).4. Store all detected groups of k-mers for later use in the supervised phase. Each group has its own statistical confidence level which can be used for further analysis. However, all groups are guaranteed to be more robust than the initial E selected.[000102] Data Representation[000103] Each microbiome sample, denoted as X, may contribute to the generation of a unique network, GX, comprising M nodes and EX edges. The nodes’ consistency across all networks may originate from the initial data processing: during an automated preliminary analysis of the data, infrequently appearing k-mers may be excluded based on a filtering criterion (e.g., only k-mers that appear at least twice in at least two samples are retained; or a stricter criterion may be used to further reduce the number of k-mers). A manageable quantity M of k-mers remain, e.g., that is conducive to our network analytics. This provision is facilitated by the power-law distribution of k-mer repetitions across the data, which guarantees that the majority of k-mers will indeed be unique and subsequently filtered out, while predicting that there will still be a substantial number of repeating k-mers recurring multiple times.[000104] In one example, k-mers may be 30-mers. The number k = 30 is large enough to allow for sufficient expressivity of the k-mers but low enough to allow different k-mers to appear in the same read, subsequently manifested as an edge in the k-mers network. From a search space of 430possible 30-mers, there are approximately one million 30-mers, translating into networks of one million nodes, a scale that is computationally feasible to handle.[000105] Note that changing the value of k trades the maximum number of possible k-mers for the maximum number of possible connections between them. Various values from k = 20 to k = 80 could be used with similar attributes, potentially producing DNA patterns that would further enhance the model’s predictive capabilities.[000106] An enhanced model that repeats the flow described here for various values of k, each resulting in a different set of k-mers, can still be managed computationally efficiently. Even a model with k-mers for 60 different values of k would result in networks of around 100 million nodes. Although such networks may initially seem too large, due to the efficient nature of embodiments of the invention, these networks can still be analyzed with practical computational resources.[000107] Network Generation from K-MersP-655574-PC[000108] The generation of each network, denoted as GX, may involve a combination of genomic sequencing and preliminary data analysis, subsequently processed by an Artificial Intelligence (AI) engine (e.g., processor(s) 112 of Fig. 1). In the initial stage, the microbial DNA is fragmented into a set of k-mers (e.g., by processor(s) 108 of Fig. 1), e.g., 30-mers, each of which serves as a potential node in the network produced in the next phase. This process results in a theoretical search space of 430k-mers.[000109] To improve the efficiency of the network analysis, a preliminary filtering step may be used. For each microbial sample, only those k-mers that repeat a predefined number of times (e.g., at least twice) in the same sample are retained, and the rest are eliminated, thereby significantly reducing the search space. The use of this filtering is feasible due to the fact that the repetition of k-mers across the data follows a power-law distribution, allowing us to significantly reduce the size of the k-mer search space (e.g., as most of the k-mers have a unique appearance throughout each sample), while still statistically guaranteeing the existence of a large enough selection of k-mers that appear multiple times in each sample. This method results in approximately one million unique 30-mers that form the consistent nodes V used across all networks.[000110] When new samples are received, they can be easily processed such that only k-mers from V are used for their analysis. From time to time, the nodes V can be refreshed by rerunning this process, expanding the set of supported k-mers and updating the internal ordering of the networks, allowing for new k-mers originating from new data to be included in the overall analysis. This process, however, is not mandatory, ensuring flexibility in our analytical approach.[000111] Once the initial construction of the set of supported k-mers is completed, each microbiome sample X can be analyzed and its corresponding network GX = G(V, EX) can be constructed, e.g., as follows:1. The sample X may contain a large number of ’’reads” (e.g., between 1 and 10 million(s)), each a string of nucleotides of length 100 to 150 bases.2. For each read, extract all k-mers that are contained in the read and filter this list of k-mers per the list of supported k-mers V.3. For each pair v1 and v2 of supported k-mers that are contained in this read, add an edge (v1, v2) to EX.[000112] Note that whereas each network GX has its unique set of edges EX, the nodes in the network may be the same group V, e.g., established during the initial filtering phase, using the multitude of available microbiome samples. In addition, the adjacency matrix of the network GX may representP-655574-PConly the existence of connections without storing their actual strengths or quantities. In other words, a Boolean representation can be used for EX, further reducing time and space complexities. Alternatively, the strengths or quantities of the connections may be stored and modeled.[000113] Once the networks GX are constructed from single microbial samples, further networks GS can be constructed on demand, for various groups S, by easily applying a logical OR operator on the adjacency matrices corresponding to the networks GX for every X∈S. Each of these networks can then be used as input for the network anomalies detection phase, allowing for the discovery of new data patterns, observable through the data-perspective provided by the semantics of the group S.[000114] Detection of Local Network Anomalies in the K-mer Networks[000115] Given a network GX(V, EX) (or GS(V, ES), the degrees of its nodes closely follows a power-law distribution: P (k) ∝ k-αLiterature shows that in many real-world networks that follows a power-law degree distribution this pattern applies degrees larger than some minimal degree dmin. Namely, for some normalization constant c, the probability of a node v to have degree d > dmin equals, e.g.:[000116] The expected value may first be calculated for the normalization constant c using values obtainable from the data:[000117] Definition 5.1. Let V* denote all nodes of degree at least dmin and let E* denote the edges that have at least one side in V*, e.g.:V* = {(v e V)\(deg(v) > dmin }E* = {((v, u) e E)\( e V*) V (w e V*)}[000118] Lemma 5.2. If:W e I, w;a,.,Proof: The sum of a graph’s G(E* E*) degrees equals twice the number of its edges:P-655574-PCTaking into account the power- law distribution of the graph’s nodes, and summing the expected number of nodes with degree d for every degree d from dmin to the maximal degree dmax implies that, e.g.:2?, X5 ‘ K Wi A OfeM •"•,;&~X5.: « 5^;;d ' [Id [ ' F O' € V,, tfe^OO ■- d} -< " j; XImplying:> E d' "'[000119] From this, the expected number of nodes may be calculated having some degree d in a network G with a power-law degree distribution:[000120] Definition 5.3. Let Vg c v* denote all nodes of degree d>dmin:l K ' d’,, "r'P-655574-PCLemma 5.4. If:[000121] The denominator is a sum that goes, e.g., from dmin to t / max. Since <7max is large, this sum may be approximated, e.g., by the integral:¥[000122] The integral can be evaluated by using the power rule for integration, e.g. as:[000123] Substituting this back into the expression for | V \ gives, e.g.:[000124] This approximation holds, e.g., for a > 2, where the integral is convergent.P-655574-PC[000125] Definition 5.5. LetXdenotethepower-law normalizing constant for the graph G(V*, E*), e.g. as:v >£$ WtiS# "" Isuch that Lemma 5.4 can be written, e.g., as:[000126] Definition 5.6. Let Ed̂⊆ E denote all the edges that touch at least one node in V. These edges may be divided into two complementing and mutually exclusive groups “internal edges” E (edges starting and ending in V^) and “external edges” Ed̂out(edges that have one side in Vd̂and one side in V\Vd̂), e.g.:[000127] Note that for every value of d the values of \E \ and |Ed̂out| can be acquired efficiently by counting edges in the network. The ratio between the number of “internal edges” and the number of “overall edges” can therefore be calculated for d, which may represent the internal connectivity for degree d.[000128] Definition 5.7. Let βd̂denote the internal connectivity of the network G for the nodes of degree d, e.g. as.>[000129] Since the number of edges in E equals the sum of degrees of the nodes inminus the edges that has their two sides in V (e.g., as these are counted twice), and since the degrees of the nodes in is exactly d, then:SA' i ss.P-655574-PC[000130] Note that the value of βd̂can be efficiently calculated from the data for every value of d. Some groups of nodes of the same degree may have high internal connectivity and some may have a lower one. It is therefore interesting to determine the expected value of the internal connectivity of various degrees, and a given internal connectivity value threshold above which would be considered “too strong” (e. g., representing a local network structure whose probability to spontaneously emerge in a network with power-law distribution of the degree is extremely low).[000131] Theorem 5.8 provides an example anomaly criterion that for every degree d greater than dmin can ensure that the set of nodes of degree d are “too anomalous” with probability (1 — e) for every small threshold e. This criterion applies for the internal connectivity of this group of nodes (e.g., the ratio between the number of edges connecting nodes of the same degree, and edges connecting nodes of different degrees), and may depend (only) on the degree d and the network structural properties a, \E*\, dmin and dmax.[000132] Theorem 5.8. If:" > 1 I'd 'then for any small threshold e and any degree d > dmin the group of nodes of degree d with internal connectivity βd̂may be considered “unlikely to spontaneously emerge” with respect to the threshold e if e.g. the following is satisfied by βd̂:[000133] Proof. For d > dmmand for an edge e 6 Eg, by definition at least one of its sides is in Pg. The probability that the second side of this edge also ends in Pg as well is affected by e.g. two factors:P-655574-PCfirst, the more nodes there are inthe greater the chance that they would be selected for the edge e. Second, as the network has a power-law degree distribution, it may be assumed that it also follows the “preferential attachment” principle (e.g., that nodes with higher degree are more likely to acquire new connections, leading to a skew in the distribution). Under this principle, the probability of a node being chosen for a new connection is proportional to its degree, and the probability for the edge e to have both sides in is therefore a Bernoulli trial with success probability, e.g.:[000134] A value of that is “too high” (e.g., unlikely to occur with a probability less than some threshold e) may indicate that the number of successes from |Ed̂| trials is greater than some threshold xthres.[000135] To calculate an upper threshold for the number of successes in Bernoulli trials that is considered statistically significant at level e, and since the number of trials N is large, and the success probability p is not too close to 0 or 1, the Binomial distribution may be approximated by a Normal distribution. If X ~ Binomial(N,p), it can be approximated as X ≈ Y where Y ~ Normal(μ, σ2) with μ = Np and σ2= Np(1 - p), e.g.:X ~ Normal (Np, Np(\ — p))[000136] This approximation allows us to express the cumulative distribution function (CDF) of X in terms of the CDF of the Normal distribution < F>, e.g., as follows:[000137] To find a threshold xthres such that the probability of observing more than xthres successes is less than some small e, the Normal approximation may be used to give, e.g.:P-655574-PC❖ hi... uvhV;: U --- rM- Vp -- -A[000138] This provides the threshold value of the number of successes for which the probability of achieving that number or more is less than e.[000139] Using the original expressions for N and p this means that the number of internal edges above which a group of nodes with a given degree is considered “unlikely” is, e.g.:\,'A" "-' d‘ uh I uo* th' *[000140] This means e.g., that:Ri uruneurJP-655574-PC[000141] Squaring both sides:ML,. - L>\; ' t owhere, e.g.:[000142] Solving this quadratic equation, and noting that A 1 > 0, yields, e.g.:[000143] Using Lemma 5.4 this can be written as, e.g.:P-655574-PC[000144] Recalling Definition 5.7 this means that for a given degree d a group of nodes would be considered “unlikely” for an internal connectivity greater than, e.g.:[000145] Note that the memory and time complexity of this approach is remarkably low due to the aforementioned techniques, which may allow for regular retraining as new data is obtained. This low complexity, coupled with the utilization of large k for k-mers (e.g., k ≥ 10, 20 or 30), enhances prediction accuracy significantly. The expressivity of large k-mers, which refers to their ability to encapsulate functionality, contributes to the robustness and precision of our analysis.[000146] Macroscopic Microbial Data Properties: This model reveals a striking and noteworthy pattern in the distribution of the repetition of the reads and the k-mers. Specifically, the expression of these repetitions follows a power-law distribution. This observation was observed to be consistent and sustained, evident in the data extracted from multiple humans, and across both read and k-mer repetitions.[000147] The emergence of this power-law distribution was observed to be data agnostic, i.e., that it is not an artifact of specific data acquisition processes or specific to the type of data being studied. Indeed, the observed power-law distribution suggests the existence of underlying principles governing these distributions, which extend beyond the specifics of this application.[000148] One might suggest that the shotgun sequencing method, which was employed to sequence the DNA samples, could somehow skew the data towards this power-law distribution. However, this is highly unlikely. While the shotgun sequencing method can indeed introduce certain biases into the data, there is no reason to suspect that it would give rise to a power-law distribution. This is because shotgunP-655574-PCsequencing, by its nature, is a random process and, as such, is not predisposed towards creating any specific pattern, much less one as distinctive as a power-law distribution.[000149] Moreover, when the data is pooled from several humans together and then analyzed the distribution of popularity, the power-law distribution is consistently observed. This further strengthens the assertion that the power-law distribution is not an artifact of data acquisition. If it were, the distribution would be different for each human, reflecting the individual variances in data collection. The consistent manifestation of the power-law distribution across different humans suggests a more profound, universal principle at work.[000150] Interestingly, the power-law distribution was observed in the repetition of k-mers, as well as whole reads. This consistency across different levels of granularity in the data adds another layer of validation to these findings. The emergence of a power-law distribution at different scales strongly suggests the existence of scale-free, fractal-like patterns in the data, which is a key characteristic of powerlaw distributions.[000151] Taken together, this strongly suggest that the observed power-law distribution isnotamere artifact of specific methodology, but rather points to a fundamental biological phenomena that has profound implications for our understanding of microbiome DNAand the factors influencing the efficacy of drugs.[000152] Variations:[000153] While this methodology attempts to mitigate the potential for bias, with the division of humans into ’’control humans”, ’’train humans”, and ’’test humans” carefully executed to take into account factors such as the age of humans, other factors may also be taken into account (e.g., by introducing stratification or covariate balancing) to reduce biases based thereon. For example, individual differences in human physiology, gender, diet, and environmental factors could be incorporated to reduce their effect onboth the composition of the microbiome and the biological condition measurements.[000154] Additionally or alternatively, instead of no overlap between the human groups from which microbiomes were sampled and those from which biological conditions was measured (which may present another potential source of bias), the groups may in some embodiments overlap. While a small microbiome sample may be representative of the entire host group’s microbiome, this assumption may not hold in all situations. Variations in individual human microbiomes within the same group could potentially impact the effectiveness of the administered drugs or treatments and thus should be sampled in both groups.P-655574-PC[000155] Additionally or alternatively, to improve the predictive framework the accuracy and volume of the labeled data during the supervised learning stage may be bolstered to improve the model’s predictive capability. For example, multiple different measurement sources may be used to cross-validate labeled data.[000156] Impact: Experimental results were promising, demonstrating a correlation between certain microbiome patterns and drug efficacy (see e.g. Fig. 8). This allows predicting the impact of drugs or treatments based on a human’s microbiome, enhancing the average effect by using the drug or treatment only when it is expected to be effective (e.g., have above-threshold efficacy or degree of biological improvement).[000157] This methodology may assume strong microbiome homogeneity across the humans in a group or the general population. This means that while the microbiome samples were collected from one group of humans, these samples are considered indicative of the microbiome composition of a broader related or general group. The validity of this assumption is confirmed when these microbiome samples are successfully used to predict drug or treatment efficacy in a distinct group of humans. This suggests that despite no overlap in the humans from which the microbiome and biological condition measurements were taken, the microbiome sample effectively represents the group’s overall microbiome. This may result in a technique that is both robust and scalable, as it allows for a comprehensive prediction model without requiring invasive and extensive sampling from every human.[000158] Embodiments of the invention establish a novel methodology for predicting the efficacy of drugs or other treatments, and a potential of using host microbiomes as a tool to predict other biologically significant outcomes based on microbiome composition, such as disease susceptibility. Such predictive models may aid in early intervention strategies and improve the overall health and productivity of the human host.[000159] Experiments indicate that microbiome samples from a small group of hosts may be representative of related groups or the general population’s microbiome. This may be used for non- invasive, rapid and cost-effective microbiome sampling techniques that could be used at scale.[000160] Given the reliance of model training on the availability and quality of data, embodiments of the invention may incorporate a continuous and broad-scale collection of microbiome and biological condition data, to further refine the predictive model, increase its accuracy, and extend its applicability across different groups or demographics of hosts.P-655574-PC[000161] Embodiments of the invention may be used to predict the response of any biological condition measurable in the microbiome sequence to drugs or treatments (e.g., any unintended consequences of their use) to understanding the holistic impact of these drugs or treatments and provide a comprehensive view and aid in making informed decisions regarding their widespread use.[000162] Improved Cryptographic Security[000163] Due to the private nature of genetic information, patient, host and donor DNA data may be encrypted to provide secure storage, secure transmission and allow distributed computations thereon across multiple servers without (or with reduced) security risk (e.g., to execute inter-server communication or reduced risk of a security breach at one of multiple servers).[000164] In some embodiments of the invention, a cryptography application may automatically encrypt the private or sensitive DNA sequences of the patients, hosts and / or donors (without human observation or intervention that would vitiate privacy). The encrypted data may then be transmitted securely over a network from a first to a second computing device without the second or either devices accessing the unencrypted data. In an embodiment, the system of Fig. 1 may provide secure cryptographic communication between a first computer terminal (e.g., genetic sequencer 102) and a second computer terminal (e.g., sequence analyzer 106) over a communication channel therebetween. The first computer terminal may receive plaintext(s) of a patient’s private microbiome DNA sequences, transform the plaintext(s) into a plurality of microbiome DNA sequences m (e.g., k-mers), encode each of the microbiome DNA sequences m under an encryption scheme (e.g., homomorphic encryption) to produce one or more ciphertext(s) ct The first computer terminal may then transmit the ciphertext(s) ct to the second computer terminal over a communication channel, without exposing the microbiome DNA sequences m themselves to the second computer terminal. The second computer terminal may execute a cryptography circuit, such as, a homomorphic encryption circuit, on the ciphertext(s) ct to analyze and operate on the microbiome DNA sequences (e.g., according to embodiments of the invention) under encryption, without exposing any of the underlining unencrypted data. The private decryption key may securely stored, e.g., only at a secure user’s (e.g., patient, host or donor’s) device and not at the first and / or second computer terminals.[000165] Conclusion[000166] Some embodiments of the invention may use “features” for training the groups in the first unsupervised learning stage. Features may be each instance of an anomalous group of microbiome DNA sequences or k-mer nodes (or score thereof, e.g., how many times the group occurs) detected in hosts.P-655574-PCTraining may generate a correlation between the feature (e.g., group or score thereof) and a biological condition (e.g., drug efficacy). Some embodiments of the invention may select only above threshold correlative strength between group (score) and biological condition.[000167] Embodiments of the invention described herein pertain to human hosts. Hosts or donors of biological samples sequenced into the microbiome DNA sequences discussed herein are living organisms. One or more processor(s) may sequence the organism’s microbiome DNA obtained from its biological sample to identify the genetic constituents of the DNA. The processor(s) used to sequence DNA may be the same processor(s) or different processor(s) (e.g., external service provider) used to analyze and model the data. In some embodiments, "shotgun" sequencing may be used (e.g., providing rich data, but with no semantic association to the microbes themselves).[000168] While microbiome DNA is described herein, embodiments of the invention encompass a host’s microbiome DNA alone, microbiome DNA combined with the host’s genomic DNA, or the host’s genomic DNA alone, for example, sequenced from any biological host sample or specimen or mixture thereof including, for example, bile, blood, feces, skin, tissue, saliva, urine, vaginal samples, stem-cells, tumors, biopsies, etc. DNA may be sampled from a single host sample or a DNA mixture sequenced from multiple hosts or multiple samples from a single host.[000169] As used herein, a “DNA” may refer to Deoxyribonucleic acid and / or Ribonucleic acid (RNA), such as, messenger RNA (mRNA), transfer RNA (tRNA), and / or ribosomal RNA (rRNA), complementary DNA (cDNA), cell-free DNA, viral genomes, fungal genomes, bacterial genomes, archaeal genomes, host-derived nucleic acids, and any other nucleic-acid-based biological sequence obtained from a biological sample, whether via metagenomic, metatranscriptomic, or other sequencing methodologies. RNA contains adenine (A), cytosine (C), guanine (G), or uracil (U) nucleotide bases, whereas DNA contains adenine (A), cytosine (C), guanine (G) and thymine (T). DNA sequence data structures may include, for example, one or more vectors, scalar values, functions, sequences, sets, matrices, tables, lists, arrays, and / or other data structures, representing biological genetic material including one or more bases, nucleotides, genes, alleles, codons or other generic material.[000170] As used herein, “therapeutic response” or “treatment response” indicates a non-neutral and / or non-random response to a medical treatment as verified in one or more physiological test(s) that may be measured e.g., once or over time. In various embodiments, therapeutic response may indicate a completely positive therapeutic benefit or a partially positive therapeutic benefit and a partially negativeP-655574-PCtherapeutic detriment, such as with side-effects. Therapeutic response may be an overall positive aggregated, e.g., where positive effects predominate over negative effects.[000171] As used herein, “drug x” may refer for example to any of the following drugs or equivalent molecule(s) and / or molecular variants with the same or equivalent operating mechanism or biological effect that may be administered to treat biological conditions of the following respective indications Y: Vedolizumab (Entyvio®) for Crohn's Disease (CD), Vedolizumab (Entyvio®) for Ulcerative Colitis (UC), Risankizumab (Skyrizi®) for Crohn's Disease (CD), Adalimumab (Humira®) for Crohn's Disease (CD), Infliximab (Remicade®) for Crohn's Disease (CD) Ustekinumab (Stelara®) for Crohn's Disease (CD), Upadacitinib (Rinvoq®) for Ulcerative Colitis (UC), and Certolizumab Pegol (Cimzia®) for Crohn's Disease (CD). Other or different combinations of drugs and indications may also be used.[000172] Variables used herein, e.g., k, N, M, Q, etc. may be integers.[000173] As used herein, “practical,” “possible,” or “finite” time, computations, problems or tasks may refer to those that can be executed with a standard computer system on the order of less than a few (e.g., one or two) hours or days, and “impractical” or “impossible” time, computations, problems or tasks may refer to those that can only be executed with a standard computer system on the order of greater than a few (e.g., one or two) hours or days.[000174] As used herein, “only” may mean exclusively, almost exclusively such as having less than 1-10% exception or with a number of exceptions that is less than a small proportion thereof; occurring in greater than 1-10% of instances or with a number of instances that is greater than a dominant proportion thereof, predominantly, in a majority (≥50%) of instances.[000175] In the foregoing description, various aspects of the present invention are described. For purposes of explanation, specific configurations and details are set forth in order to provide a thorough understanding of the present invention. However, it will also be apparent to persons of ordinary skill in the art that the present invention may be practiced without the specific details presented herein. Furthermore, well known features may be omitted or simplified in order not to obscure the present invention.[000176] Unless specifically stated otherwise, it is appreciated that throughout the specification discussions utilizing terms such as "processing," "computing," "calculating," "determining," or the like, refer to the action and / or processes of a computer or computing system, or similar electronic computing device, that manipulates and / or transforms data represented as physical, such as electronic, quantitiesP-655574-PCwithin the computing system's registers and / or memories into other data similarly represented as physical quantities within the computing system's memories, registers or other such information storage, transmission or display devices.[000177] Embodiments of the invention may include an article such as a computer or processor readable non-transitory storage medium, such as for example a memory, a disk drive, or a USB flash memory device (e.g., memory unit(s) 114 and / or 118 of Fig. 1) encoding, including or storing instructions, e.g., computer-executable instructions, which when executed by a processor or controller (e.g., controller(s) or processor(s) 108 and / or 112 of Fig. 1), cause the processor or controller to carry out methods disclosed herein.[000178] Different embodiments are disclosed herein. Features of certain embodiments may be combined with features of other embodiments; thus certain embodiments may be combinations of features of multiple embodiments.[000179] The foregoing description of the embodiments of the invention has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. It should be appreciated by persons of ordinary skill in the art that many modifications, variations, substitutions, changes, and equivalents are possible in light of the above teaching. For example, it should be appreciated that sign conventions are equivalent and that embodiments of the invention in which values are above a lower bound or threshold are equivalent to embodiments of the invention in which values are below an upper bound or threshold, since the difference is a mere convention of sign. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the invention.

Claims

P-655574-PCCLAIMS1. A method for detecting biomarkers in microbiome DNA of one or more human hosts to predict a biological condition, the method comprising:storing a plurality of microbiome DNA sequences sequenced from multiple microbial organisms in microbiome DNA of one or more human hosts;generating a microbiome network comprising a plurality of nodes representing the respective plurality of microbiome DNA sequences and a plurality of edges representing a cooccurrence of each pair of microbiome DNA sequences in a same sample or sub-length of the microbiome DNA;bundling, into distinct groups, microbiome DNA sequences represented by nodes based on the node’s degree, the degree of each node quantifying a number of edges that contain the node indicating a number of unique microbiome DNA sequences that co-occur with the microbiome DNA sequence represented by the node in the same sample or sub-length of the microbiome DNA;validating groups of microbiome DNA sequences that each have an internal connectivity between nodes of the same group that satisfies an anomaly condition indicating the microbiome DNA sequences of that group co-occur with a probability that is unlikely randomly statistical;generating a training dataset correlating the validated groups of microbiome DNA sequences from one or more human hosts with a biological condition measured in one or more human hosts; andtraining a machine learning model with the training dataset to input microbiome DNA sequences from one or more human hosts and predict the biological condition for one or more human hosts.

2. The method of claim 1 comprising:storing a plurality of microbiome DNA sequences sequenced from multiple microbial organisms in microbiome DNA of a human patient;predicting the human patient has the biological condition in response to the machine learning model inputting the plurality of microbiome DNA sequences from the human patient comprising one or more of the validated groups of microbiome DNA sequences correlated with the biological condition measured in the one or more human hosts; andadministering a treatment for the biological condition to the human patient predicted to have the biological condition.P-655574-PC3. The method of claim 1, wherein the internal connectivity is determined based on a ratio between a number of internal edges connecting two nodes internal to the same group and a total number of overall edges connecting nodes within the group to any other node internal or external to the group.

4. The method of claim 3, wherein the internal connectivity condition for a group of nodes having a same degree d is that the group’s internal connectivity is greater than or equal to:'if:>, where xthresis a threshold in which the probability of observing more than the threshold is less than a value e and |Vd|is the number of nodes in the microbiome network of degree d.

5. The method of claim 1, wherein the internal connectivity is determined based on a number of nodes in the same group, number of edges in the same group, a number of pre-defined partially or fully connected clusters in the same group or deviation thereof.

6. The method of claim 1, wherein the internal connectivity condition is satisfied if the internal connectivity for a group of nodes deviates from an expected internal connectivity in a network that follows a power law distribution.

7. The method of claim 1 comprising filtering the plurality of microbiome DNA sequences to include only microbiome DNA sequences that occur greater than a predefined integer number of times in the same sample or sub-length of the microbiome DNA.

8. The method of claim 1 comprising filtering the plurality of microbiome DNA sequences to include only nodes within a predefined degree range and edges connecting those nodes.

9. The method of claim 1 comprising super-positioning a plurality of networks each representing a distinct microbiome DNA sample to generate a composite network representing a plurality of microbiome DNA samples.

10. The method of claim 1 comprising retraining the model based on new microbiome DNA sequenced from multiple microbial organisms in one or more new microbiome samples of the same or different one or more human hosts.

11. The method of claim 1, wherein the biological condition of the one or more human hosts is selected from the group consisting of: medicinal efficacy, gastrointestinal or overall health,P-655574-PCdisease susceptibility, disease tolerance, likelihood of disease recovery, life expectancy, and / or fatality risk.

12. The method of claim 1, wherein the microbiome DNA is sequenced from a biological sample selected from the group consisting of: feces, skin, saliva, vaginal samples, and another site in the human body.

13. A system for detecting biomarkers in microbiome DNA of one or more human hosts to predict a biological condition, the system comprising:one or more memories configured to store a plurality of microbiome DNA sequences sequenced from multiple microbial organisms in microbiome DNA of one or more human hosts; andone or more processors configured to:generate a microbiome network comprising a plurality of nodes representing the respective plurality of microbiome DNA sequences and a plurality of edges representing a co-occurrence of each pair of microbiome DNA sequences in a same sample or sub-length of the microbiome DNA,bundle, into distinct groups, microbiome DNA sequences represented by nodes based on the node’s degree, the degree of each node quantifying a number of edges that contain the node indicating a number of unique microbiome DNA sequences that co-occur with the microbiome DNA sequence represented by the node in the same sample or sub-length of the microbiome DNA,validate groups of microbiome DNA sequences that each have an internal connectivity between nodes of the same group that satisfies an anomaly condition indicating the microbiome DNA sequences of that group co-occur with a probability that is unlikely randomly statistical,generate a training dataset correlating the validated groups of microbiome DNA sequences from one or more human hosts with a biological condition measured in one or more human hosts, andtrain a machine learning model with the training dataset to input microbiome DNA sequences from one or more human hosts and predict the biological condition for one or more human hosts.

14. The system of claim 13, wherein the one or more processors are configured to:P-655574-PCstore a plurality of microbiome DNA sequences sequenced from multiple microbial organisms in microbiome DNA of a human patient, andpredict the human patient has the biological condition in response to the machine learning model inputting the plurality of microbiome DNA sequences from the human patient comprising one or more of the validated groups of microbiome DNA sequences correlated with the biological condition measured in the one or more human hosts, wherein the treatment for the biological condition is administered to the human patient predicted to have the biological condition.

15. The system of claim 13, wherein the one or more processors are configured to determine the internal connectivity based on a ratio between a number of internal edges connecting two nodes internal to the same group and a total number of overall edges connecting nodes within the group to any other node internal or external to the group.

16. The system of claim 15, wherein the one or more processors are configured to determine that a group of nodes having a same degree d satisfy the internal connectivity condition if the group’s internal connectivity is greater than or equal to:, where xthresis a threshold in which the probability of observing more than the threshold is less than a value e and |Vd|is the number of nodes in the microbiome network of degree d.

17. The system of claim 13, wherein the one or more processors are configured to determine the internal connectivity based on a number of nodes in the same group, number of edges in the same group, a number of pre-defined partially or fully connected clusters in the same group or deviation thereof.

18. The system of claim 13, wherein the one or more processors are configured to determine that the internal connectivity condition is satisfied if the internal connectivity for a group of nodes deviates from an expected internal connectivity in a network that follows a power law distribution.

19. The system of claim 13, wherein the one or more processors are configured to filter the plurality of microbiome DNA sequences to include only microbiome DNA sequences that occur greater than a predefined integer number of times in the same sample or sub-length of the microbiome DNA.P-655574-PC20. The system of claim 13, wherein the one or more processors are configured to filter the plurality of microbiome DNA sequences to include only nodes within a predefined degree range and edges connecting those nodes.

21. The system of claim 13, wherein the one or more processors are configured to super-position a plurality of networks each representing a distinct microbiome DNA sample to generate a composite network representing a plurality of microbiome DNA samples.

22. The system of claim 13, wherein the one or more processors are configured to retrain the model based on new microbiome DNA sequenced from multiple microbial organisms in one or more new microbiome samples of the same or different one or more human hosts.

23. The system of claim 13, wherein the biological condition of the one or more human hosts is selected from the group consisting of: medicinal efficacy, gastrointestinal or overall health, disease susceptibility, disease tolerance, likelihood of disease recovery, life expectancy, and / or fatality risk.

24. The system of claim 13, wherein the microbiome DNA is sequenced from a biological sample selected from the group consisting of: feces, skin, saliva, vaginal samples, and another site in the human body.

Images