SEARCH · PubMed Health

Results for “machine learning prediction”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 487 records · Page 27Linked to original sources

PubMed2026

3D epigenome of glial cell types in developing human cortex.

The human cortex is complex and heterogeneous, undergoing extensive expansion during development1,2. Our prior study of neurogenesis, including radial glia (RG), intermediate progenitor cells, excitatory neurons and interneurons demonstrated that chromatin looping underlies transcriptional regulation for lineage-specific genes, shedding light on how non-coding genetic variants contribute to neuropsychiatric disorders by means of cell-type-specific gene regulation3. RG have a crucial role in generating cellular diversity through both neurogenesis and gliogenesis and can be further classified into ventricular RG (vRG) and outer RG (oRG)4,5. Given their significance in cortical development, we conducted a comprehensive three-dimensional (3D) epigenomic analysis of four main glial populations, including vRG, oRG, oligodendrocyte precursor cells and microglia, from the mid-gestational human neocortex. By integrating gene expression, chromatin accessibility, DNA methylation and 3D chromatin interactions, we identified cell-type-specific candidate cis-regulatory elements (cCREs) and validated their regulatory function using transgenic mouse embryos. Using machine learning, we prioritized 112 schizophrenia risk variants within glia cCREs and further confirmed the predicted vRG enhancer disruption by the rs4449074 risk allele in vivo. Finally, oRG cCREs are enriched for human accelerated regions compared with other cCREs and a subset of human accelerated regions show activity differences from their chimpanzee orthologues that interact with genes involved in neuronal development. Our findings advance the understanding of human-specific gene regulation during corticogenesis.

Journal Article↗

PubMed2025

The signed two-space proximity model for learning representations in protein-protein interaction networks.

MOTIVATION: Accurately predicting complex protein-protein interactions (PPIs) is crucial for decoding biological processes, from cellular functioning to disease mechanisms. However, experimental methods for determining PPIs are computationally expensive. Thus, attention has been recently drawn to machine learning approaches. Furthermore, insufficient effort has been made toward analyzing signed PPI networks, which capture both activating (positive) and inhibitory (negative) interactions. To accurately represent biological relationships, we present the Signed Two-Space Proximity Model (S2-SPM) for signed PPI networks, which explicitly incorporates both types of interactions, reflecting the complex regulatory mechanisms within biological systems. This is achieved by leveraging two independent latent spaces to differentiate between positive and negative interactions while representing protein similarity through proximity in these spaces. Our approach also enables the identification of archetypes representing extreme protein profiles. RESULTS: S2-SPM's superior performance in predicting the presence and sign of interactions in SPPI networks is demonstrated in link prediction tasks against relevant baseline methods. Additionally, the biological prevalence of the identified archetypes is confirmed by an enrichment analysis of Gene Ontology (GO) terms, which reveals that distinct biological tasks are associated with archetypal groups formed by both interactions. This study is also validated regarding statistical significance and sensitivity analysis, providing insights into the functional roles of different interaction types. Finally, the robustness and consistency of the extracted archetype structures are confirmed using the Bayesian Normalized Mutual Information (BNMI) metric, proving the model's reliability in capturing meaningful SPPI patterns. AVAILABILITY: S2-SPM is implemented and freely available under the MIT license at https://github.com/Nicknakis/S2SPM.

Protein Interaction Mapping↗

PubMed2026

Machine learning-guided risk stratification in elderly AML based on genomic, immunophenotypic and therapeutic profiles.

BACKGROUND: Elderly patients with acute myeloid leukemia (AML) exhibit considerable biological and clinical heterogeneity, hindering precise prognosis. Existing prognostic systems inadequately capture the complexity of elderly AML due to their reliance on data from younger cohorts and omission of key factors like immunophenotypic markers and therapeutic profiles. This study aimed to develop and internally validate a machine learning-based prognostic model specifically tailored to elderly AML patients. METHODS: A total of 156 patients were analyzed using a two-stage modeling strategy. Clinical and genomic variables were modeled first, followed by independent analysis of immunophenotypic features. Feature selection was performed using multilayer perceptron (MLP) and random forest (RF), while multivariate Cox regression was used for final model construction. Internal validation was conducted using 1000 bootstrap iterations to assess model stability and performance. RESULTS: The model demonstrated strong predictive performance, with a concordance index (C-index) of 0.702. Time-dependent area under the curve (AUC) and calibration plots confirmed accurate prediction of 1-, 3-, and 5-year overall survival. Decision curve analysis indicated favorable net benefit across a range of threshold probabilities. Key independent prognostic factors identified included TP53 mutations, high CD13 expression, and IDH2 mutations. CONCLUSION: This model provides a robust and interpretable tool for individualized risk stratification in elderly AML. By integrating genomic, immunophenotypic, and therapeutic variables, it may help optimize treatment decisions and improve outcomes for this vulnerable population. Future efforts should focus on external validation and integration of dynamic biomarkers.

Humans↗

PubMed2006

Prediction of RNA binding sites in proteins from amino acid sequence.

RNA-protein interactions are vitally important in a wide range of biological processes, including regulation of gene expression, protein synthesis, and replication and assembly of many viruses. We have developed a computational tool for predicting which amino acids of an RNA binding protein participate in RNA-protein interactions, using only the protein sequence as input. RNABindR was developed using machine learning on a validated nonredundant data set of interfaces from known RNA-protein complexes in the Protein Data Bank. It generates a classifier that captures primary sequence signals sufficient for predicting which amino acids in a given protein are located in the RNA-protein interface. In leave-one-out cross-validation experiments, RNABindR identifies interface residues with >85% overall accuracy. It can be calibrated by the user to obtain either high specificity or high sensitivity for interface residues. RNABindR, implementing a Naive Bayes classifier, performs as well as a more complex neural network classifier (to our knowledge, the only previously published sequence-based method for RNA binding site prediction) and offers the advantages of speed, simplicity and interpretability of results. RNABindR predictions on the human telomerase protein hTERT are in good agreement with experimental data. The availability of computational tools for predicting which residues in an RNA binding protein are likely to contact RNA should facilitate design of experiments to directly test RNA binding function and contribute to our understanding of the diversity, mechanisms, and regulation of RNA-protein complexes in biological systems. (RNABindR is available as a Web tool from http://bindr.gdcb.iastate.edu.).

Amino Acid Motifs↗

PubMed2005

Boosting alternating decision trees modeling of disease trait information.

We applied the alternating decision trees (ADTrees) method to the last 3 replicates from the Aipotu, Danacca, Karangar, and NYC populations in the Problem 2 simulated Genetic Analysis Workshop dataset. Using information from the 12 binary phenotypes and sex as input and Kofendrerd Personality Disorder disease status as the outcome of ADTrees-based classifiers, we obtained a new quantitative trait based on average prediction scores, which was then used for genome-wide quantitative trait linkage (QTL) analysis. ADTrees are machine learning methods that combine boosting and decision trees algorithms to generate smaller and easier-to-interpret classification rules. In this application, we compared four modeling strategies from the combinations of two boosting iterations (log or exponential loss functions) coupled with two choices of tree generation types (a full alternating decision tree or a classic boosting decision tree). These four different strategies were applied to the founders in each population to construct four classifiers, which were then applied to each study participant. To compute average prediction score for each subject with a specific trait profile, such a process was repeated with 10 runs of 10-fold cross validation, and standardized prediction scores obtained from the 10 runs were averaged and used in subsequent expectation-maximization Haseman-Elston QTL analyses (implemented in GENEHUNTER) with the approximate 900 SNPs in Hardy-Weinberg equilibrium provided for each population. Our QTL analyses on the basis of four models (a full alternating decision tree and a classic boosting decision tree paired with either log or exponential loss function) detected evidence for linkage (Z >or= 1.96, p < 0.01) on chromosomes 1, 3, 5, and 9. Moreover, using average iteration and abundance scores for the 12 phenotypes and sex as their relevancy measurements, we found all relevant phenotypes for all four populations except phenotype b for the Karangar population, with suggested subgroup structure consistent with latent traits used in the model. In conclusion, our findings suggest that the ADTrees method may offer a more accurate representation of the disease status that allows for better detection of linkage evidence.

Algorithms↗

PubMed2003

Optimization of neural network architecture using genetic programming improves detection and modeling of gene-gene interactions in studies of human diseases.

BACKGROUND: Appropriate definition of neural network architecture prior to data analysis is crucial for successful data mining. This can be challenging when the underlying model of the data is unknown. The goal of this study was to determine whether optimizing neural network architecture using genetic programming as a machine learning strategy would improve the ability of neural networks to model and detect nonlinear interactions among genes in studies of common human diseases. RESULTS: Using simulated data, we show that a genetic programming optimized neural network approach is able to model gene-gene interactions as well as a traditional back propagation neural network. Furthermore, the genetic programming optimized neural network is better than the traditional back propagation neural network approach in terms of predictive ability and power to detect gene-gene interactions when non-functional polymorphisms are present. CONCLUSION: This study suggests that a machine learning strategy for optimizing neural network architecture may be preferable to traditional trial-and-error approaches for the identification and characterization of gene-gene interactions in common, complex human diseases.

Algorithms↗

PubMed2005

Classifying spatial patterns of brain activity with machine learning methods: application to lie detection.

Patterns of brain activity during deception have recently been characterized with fMRI on the multi-subject average group level. The clinical value of fMRI in lie detection will be determined by the ability to detect deception in individual subjects, rather than group averages. High-dimensional non-linear pattern classification methods applied to functional magnetic resonance (fMRI) images were used to discriminate between the spatial patterns of brain activity associated with lie and truth. In 22 participants performing a forced-choice deception task, 99% of the true and false responses were discriminated correctly. Predictive accuracy, assessed by cross-validation in participants not included in training, was 88%. The results demonstrate the potential of non-linear machine learning techniques in lie detection and other possible clinical applications of fMRI in individual subjects, and indicate that accurate clinical tests could be based on measurements of brain function with fMRI.

Adult↗

PubMed2022

Flnc: Machine Learning Improves the Identification of Novel Long Noncoding RNAs from Stand-Alone RNA-Seq Data.

Long noncoding RNAs (lncRNAs) play critical regulatory roles in human development and disease. Although there are over 100,000 samples with available RNA sequencing (RNA-seq) data, many lncRNAs have yet to be annotated. The conventional approach to identifying novel lncRNAs from RNA-seq data is to find transcripts without coding potential but this approach has a false discovery rate of 30-75%. Other existing methods either identify only multi-exon lncRNAs, missing single-exon lncRNAs, or require transcriptional initiation profiling data (such as H3K4me3 ChIP-seq data), which is unavailable for many samples with RNA-seq data. Because of these limitations, current methods cannot accurately identify novel lncRNAs from existing RNA-seq data. To address this problem, we have developed software, Flnc, to accurately identify both novel and annotated full-length lncRNAs, including single-exon lncRNAs, directly from RNA-seq data without requiring transcriptional initiation profiles. Flnc integrates machine learning models built by incorporating four types of features: transcript length, promoter signature, multiple exons, and genomic location. Flnc achieves state-of-the-art prediction power with an AUROC score over 0.92. Flnc significantly improves the prediction accuracy from less than 50% using the conventional approach to over 85%. Flnc is available via GitHub platform.

RNA-seq↗

PubMed1994

Improving prediction of preterm birth using a new classification scheme and rule induction.

Prediction of preterm birth is a poorly understood domain. The existing manual methods of assessment of preterm birth are 17%-38% accurate. The machine learning system LERS was used for three different datasets about pregnant women. Rules induced by LERS were used in conjunction with a classification scheme of LERS, based on "bucket brigade algorithm" of genetic algorithms and enhanced by partial matching. The resulting prediction of preterm birth in new, unseen cases is much more accurate (68%-90%).

Algorithms↗

PubMed1994

On the use of machine learning to identify topological rules in the packing of beta-strands.

The machine learning program GOLEM was applied to discover topological rules in the packing of beta-sheets in alpha/beta-domain proteins. Rules (constraints) were determined for four features of beta-sheet packing: (i) whether a beta-strand is at an edge; (ii) whether two consecutive beta-strands pack parallel or anti-parallel; (iii) whether two beta-strands pack adjacently; and (iv) the winding direction of two consecutive beta-strands. Rules were found with high predictive accuracy and coverage. The errors were generally associated with complications in domain folds, especially in one doubly would domains. Investigation of the rules revealed interesting patterns, some of which were known previously, others that are novel. Novel features include (i) the relationship between pairs of sequential strands is in general one of decreasing size; (ii) more sequential pairs of strands wind in the direction out than in; and (iii) it takes a larger alteration in hydrophobicity to change a strand from winding in the direction out than in. These patterns in the data may be the result of folding pathways in the domains. The rules found are of predictive value and could be used in the combinatorial prediction of protein structure, or as a general test of model structures, e.g. those produced by threading. We conclude that machine learning has a useful role in the analysis of protein structures.

Amino Acid Sequence↗

PubMed2005

Flow rate of some pharmaceutical diluents through die-orifices relevant to mini-tableting.

The effects of cylindrical orifice length and diameter on the flow rate of three commonly used pharmaceutical direct compression diluents (lactose, dibasic calcium phosphate dihydrate and pregelatinised starch) were investigated, besides the powder particle characteristics (particle size, aspect ratio, roundness and convexity) and the packing properties (true, bulk and tapped density). Flow rate was determined for three different sieve fractions through a series of miniature tableting dies of different orifice diameter (0.4, 0.3 and 0.2 cm) and thickness (1.5, 1.0 and 0.5 cm). It was found that flow rate decreased with the increase of the orifice length for the small diameter (0.2 cm) but for the large diameter (0.4 cm) was increased with the orifice length (die thickness). Flow rate changes with the orifice length are attributed to the flow regime (transitional arch formation) and possible alterations in the position of the free flowing zone caused by pressure gradients arising from the flow of self-entrained air, both above the entrance in the die orifice and across it. Modelling by the conventional Jones-Pilpel non-linear equation and by two machine learning algorithms (lazy learning, LL, and feed-forward back-propagation, FBP) was applied and predictive performance of the fitted models was compared. It was found that both FBP and LL algorithms have significantly higher predictive performance than the Jones-Pilpel non-linear equation, because they account both dimensions of the cylindrical die opening (diameter and length). The automatic relevance determination for FBP revealed that orifice length is the third most influential variable after the orifice diameter and particle size, followed by the bulk density, the difference between bulk and tapped densities and the particle convexity.

Algorithms↗

PubMed2025

Quantifying and improving rheumatoid arthritis algorithm performance in biobank settings.

OBJECTIVE: To quantify and improve the performance of standard rheumatoid arthritis (RA) algorithms in a biobank setting. METHODS: This retrospective cohort study within the Mayo Clinic (MC) Biobank and MC Tapestry Study identified RA cases by presence of at least two RA codes OR positive anti-cyclic citrullinated peptide antibodies (CCP) plus disease-modifying anti-rheumatic drug (DMARD) prescription as of 7/18/2022. Rheumatology physicians manually verified all RA cases using RA criteria and/or rheumatology physician diagnosis plus DMARD use. All other biobank participants served as non-RA controls. We defined seropositivity as rheumatoid factor and/or anti-CCP positivity. We assessed rules-based and Electronic Medical Records and Genomics (eMERGE) RA algorithms using positive predictive value (PPV). Finally, we developed a novel RA algorithm using a LASSO-based machine learning approach with five-fold cross validation. RESULTS: We identified 1,316 confirmed RA cases (968 MC Biobank, 348 Tapestry, 70 % seropositive) and 82,123 non-RA controls (mean age 65, 61 % female). The PPV of 3 RA codes was 43 %, codes plus DMARD was 54 %, and codes plus DMARD plus seropositivity was 85 %. The PPV of eMERGE was 77 %. Available in the MC Biobank, self-reported RA (PPV 10 %) only minimally improved algorithm performance (PPV from 83 % to 85 %), whereas family history of RA (PPV 3 %) worsened performance. At 90 % PPV, the novel RA algorithm incorporating key variables such as anti-CCP and DMARD use increased sensitivity by 4-11 % compared to eMERGE. CONCLUSION: Rules-based and eMERGE RA algorithms had worse performance in biobank than administrative settings. Our novel RA algorithm outperformed these standard algorithms.

Humans↗

PubMed2026

Comprehensive in silico genomics analysis of global trends and host-specific emergence of aminoglycoside resistance in Staphylococcus aureus: a One-Health perspective.

BACKGROUND: Aminoglycosides remain clinically valuable against Staphylococcus aureus. Aminoglycoside resistance in S. aureus represents a critical One Health concern and is primarily driven by aminoglycoside-modifying enzymes (AMEs), which are frequently plasmid-encoded. Although regional studies have provided valuable insights, the global epidemiology of aminoglycoside resistance determinants remains poorly characterized because comprehensive data integrating human, animal, and environmental reservoirs are still lacking. This study addresses this gap by analyzing over 110,000 S. aureus genomes (2000-2025) to map the global resistome, quantify temporal and host-specific trends, and assess the association between genetic determinants and phenotypic resistance. METHODS: We performed a retrospective One Health meta-analysis of 110,309 S. aureus genomes collected between 2000 and 2025 from 128 countries. Genomes were quality-filtered and aminoglycoside resistance determinants were identified using NCBI AMRFinderPlus (v4.0.23). Multilocus sequence typing and host-source harmonization (Human, Animal, Environment, Unknown) enabled clonal and reservoir stratification. Temporal trends in gene prevalence and resistance burden were modeled with robust regression. Geographic and host-associated structuring of key genes was assessed via χ2 and enrichment tests. Machine-learning models (elastic-net, random forests, XGBoost) were benchmarked for minimum inhibitory concentration (MIC) prediction via nested cross-validation, with performance evaluated by mean absolute error, RMSE, and SHAP-based feature importance. All analyses were conducted in R and Python using publicly available, de-identified genomic data. RESULTS: Aminoglycoside resistance-associated genes were dominated by modifying enzyme determinants, with ant(6)-Ia, ant(9)-Ia, aph(3')-IIIa, sat4, aadD1, and aac(6')-Ie/aph(2'')-Ia occurring in 14-22% of isolates worldwide. Temporal analysis revealed significant declines in several major determinants, most notably ant(9)-Ia (-2.22 percentage points per year, p < 0.001), whereas apmA exhibited a non-significant decreasing trend in animal isolates. Host structuring was marked: human clinical isolates concentrated common determinants, while animal and environmental isolates harbored rare alleles (apmA, spw, str, spd). Geographic mapping confirmed near-universal distribution of common genes but focal restriction of rare ones. Publicly available phenotypic data indicated strong activity of amikacin, whereas gentamicin showed a distinct resistant subpopulation that closely corresponded with AME gene carriage. Genotype-phenotype analyses demonstrated strong concordance, with gene-rich complements predicting resistant MIC strata and absence of determinants predicting susceptibility. Analysis across different gene classes revealed frequent co-occurrence of aminoglycoside resistance genes with determinants from other classes, such as mecA, blaZ, and MLS_B, embedding them within multidrug-resistant (MDR) genomic contexts. CONCLUSION: Over 25 years, the prevalence of aminoglycoside resistance-associated genes in S. aureus has declined for several common determinants, while rare veterinary-linked alleles are emerging in animal isolates. Strong genotype-phenotype concordance supports genomic prediction for gentamicin and amikacin, where MIC data are available, although phenotypic confirmation remains essential. The frequent co-occurrence of aminoglycoside resistance genes with other antimicrobial resistance determinants indicates their integration within co-occurrence patterns of MDR genes, defined here as clusters of co-occurring resistance genes often carried on shared mobile genetic elements. These patterns highlight the need for integrated One Health surveillance combining clinical, veterinary, and environmental monitoring with plasmid-context resolution to anticipate emerging threats.

Aminoglycosides↗

PubMed2006

[Rule induction algorithm for brain glioma using support vector machine].

A new proposed data mining technique, support vector machine (SVM), is used to predict the degree of malignancy in brain glioma. Based on statistical learning theory, SVM realizes the principle of data dependent structure risk minimization, so it can depress the overfitting with better generalization performance, since the prediction in medical diagnosis often deals with a small sample. SVM based rule induction algorithm is implemented in comparison with other data mining techniques such as artificial neural networks, rule induction algorithm and fuzzy rule extraction algorithm based on fuzzy max-min neural networks (FRE-FMMNN) proposed recently. Computation results by 10 fold cross validation method show that SVM can get higher prediction accuracy than artificial neural networks and FRE-FMMNN, which implies SVM can get higher accuracy and more reliability. On the whole data sets, SVM gets one rule with the classification accuracy of 89.29%, while FRE-FMMNN gets two rules of 84. 64%, in which the rule got by SVM is of quantity relation and contains more information than the two rules by FRE-FMMNN. All the above show SVM is a potential algorithm for the medical diagnosis such as the prediction of the degree of malignancy in brain glioma.

Algorithms↗

PubMed2026

Identification and Classification of Expressed Orphan Genes, Spurious Orphan Genes, and Conserved Genes in the Human Gut Microbiome.

Orphan genes (OGs)-genes lacking detectable homologs outside a species-are widespread in microbial genomes and are thought to contribute to their adaptation and molecular innovation. However, not all predicted OGs may represent novel functional coding sequences. False positive OGs, also called spurious OGs, can arise from gene prediction errors. We reason that OGs lacking detectable expression are more likely to be spurious. To test this, we combined large-scale metatranscriptomic profiling of the human gut microbiome with machine learning to distinguish expressed OGs from spurious ones and compare them with conserved genes (CGs) found in multiple species. Using nearly 5,000 metatranscriptome libraries, we identified ∼218,000 OGs supported by expression evidence, while ∼330,000 predicted OGs lacked detectable expression and were classified as spurious. We extracted 154 features for sequence, structural, and evolutionary properties for each gene and trained XGBoost classifiers while accounting for genomic representation. The models achieved an area under the receiver operating characteristic curve (AUC) of 0.82 in distinguishing expressed OGs from spurious OGs and an AUC of 0.93 in distinguishing expressed OGs from CGs. Interpretation based on SHAP (SHapley Additive exPlanations) revealed clear biological signals. Particularly, expressed orphans were present in more genomes than spurious ones, and expressed OGs were shorter than CGs. This work improves OG discovery and suggests that expressed OGs differ systematically from CGs and spurious OGs in sequence composition, structural constraints, and evolutionary signals.

Humans↗

PubMed2001

Multi-class protein fold recognition using support vector machines and neural networks.

MOTIVATION: Protein fold recognition is an important approach to structure discovery without relying on sequence similarity. We study this approach with new multi-class classification methods and examined many issues important for a practical recognition system. RESULTS: Most current discriminative methods for protein fold prediction use the one-against-others method, which has the well-known 'False Positives' problem. We investigated two new methods: the unique one-against-others and the all-against-all methods. Both improve prediction accuracy by 14-110% on a dataset containing 27 SCOP folds. We used the Support Vector Machine (SVM) and the Neural Network (NN) learning methods as base classifiers. SVMs converges fast and leads to high accuracy. When scores of multiple parameter datasets are combined, majority voting reduces noise and increases recognition accuracy. We examined many issues involved with large number of classes, including dependencies of prediction accuracy on the number of folds and on the number of representatives in a fold. Overall, recognition systems achieve 56% fold prediction accuracy on a protein test dataset, where most of the proteins have below 25% sequence identity with the proteins used in training.

Discriminant Analysis↗

PubMed2001

Feature mining and predictive model construction from severe trauma patient's data.

In management of severe trauma patients, trauma surgeons need to decide which patients are eligible for damage control. Such decision may be supported by utilizing models that predict the patient's outcome. The study described in this paper investigates the possibility to construct patient outcome prediction models from retrospective patient's data at the end of initial damage control surgery by using feature mining and machine learning techniques. As the data used comprises rather excessive number of features, special attention was paid to the problem of selecting only the most relevant features. We show that a small subset of features may carry enough information to construct reasonably accurate prognostic models. Furthermore, the techniques used in our study identified two factors, namely the pH value when admitted to ICU and the worst partial active thromboplastin time, to be of highest importance for prediction. This finding is pathophysiologically reasonable and represents two of three major problems with severe trauma patients, metabolic acidosis, hypothermia, and coagulopathy.

Algorithms↗

PubMed2025

Integrating explainable artificial intelligence with multiomics systems biology and electronic health record data mining for personalized drug repurposing in Alzheimer's disease.

Alzheimer's disease (AD) is characterized by region- and patient-specific molecular heterogeneity, which hinders therapeutic design. In this study, we introduce PRISM-ML (PRecision-medicine using Interpretable Systems and Multiomics with Machine Learning), an open-source integrated analysis pipeline that combines interpretable machine learning with systems biology and electronic health records data mining to elucidate the molecular diversity of AD and predict promising drug repurposing opportunities. First, we integrated and harmonized transcriptomic (bulk RNA-seq) and genomic (genome-wide association study) data from 2105 brain samples, each with matched data from the same individual (1363 AD patients, 742 controls; 9 tissues), sourced from three independent studies. Random forest classifiers with SHapley Additive exPlanations identified patient-specific biomarkers; unsupervised clustering resolved 36 molecularly distinct subtissues (defined as clusters of samples within a brain tissue that share a specific expression pattern); and gene-gene coexpression networks prioritized 262 high-centrality bottleneck genes as putative regulators of dysregulated pathways. Next, knowledge graph-based drug repurposing predicted six Food and Drug Administration (FDA)-approved drugs that simultaneously target multiple bottleneck genes and multiple AD-relevant pathways. Notably, in a large US de-identified insurance-claims database (n = 364 733), exposure to promethazine, one of the candidate drugs, was associated with a 57%-62% lower incidence of AD versus an active antihistamine comparator (adjusted hazard ratio 0.38; inverse-probability weighted 0.43; both P < .001), providing real-world support for its repurposing potential. In summary, PRISM-ML, as an explainable multiomics analysis pipeline, is readily transferable to other complex diseases, advancing precision medicine.

Alzheimer Disease↗