SEARCH · Search PubMed

Results for “Machine learning model”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

SourceResearch connectionFrom yearThrough year

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

At least 307 records · Page 17Linked to original sources

PubMed2025

Integration of single cell multiomics data by deep transfer hypergraph neural network.

Multi-omics characterization of individual cells offers remarkable potential for analyzing the dynamics and relationships of gene regulatory states across millions of cells. How to integrate multimodal data is an open problem, existing integration methods struggle with accuracy and modality-specific biological variation retention. In this paper, we present scHyper (scalable, interpretable machine learning for single cell integration), a low-code and data-efficient deep transfer model designed for integrating paired and unpaired single-cell multimodal data. We benchmark scHyper against datasets from different multimodal data. ScHyper learns a low-dimensional representation and aligns the covariance matrices of the measured modalities, achieving high accuracy even with large scale atlas-level datasets with low memory and computational time across different cell lines, shedding light on regulatory relationships between different types of omics. Altogether, we show that scHyper is a versatile and robust tool for cell-type label transfer and integration from multimodal single-cell datasets.

Single-Cell Analysis↗

PubMed2025

Opportunities for machine learning to predict cross-neutralization in FMDV serotype O.

Accurately estimating cross-neutralization between serotype O foot-and-mouth disease viruses (FMDVs) is critical for guiding vaccine selection and disease management. In this study, we developed a machine learning approach to estimate r1 values-an established measure of antigenic similarity-using VP1 sequence data and published virus neutralization titer (VNT) results. Our dataset comprised 108 serum-virus pairs representing 73 distinct FMDV strains. We applied Boruta feature selection and random forest classifiers, optimizing model performance through tenfold cross-validation and sub-sampling to address class imbalance. Predictors included pairwise amino acid distances, site-specific polymorphisms, and differences in potential N-glycosylation sites. Using a 0.3 r1 threshold to define cross-neutralization, the final model achieved high accuracy (0.96), sensitivity (0.93), and specificity (0.96) in training, and performed robustly on independent test sets - accuracy was 0.75 (95% CI 0.60 and 0.90), F1 score 0.86% and PPV 0.77. Importantly, key VP1 residues-positions 48, 100, 135, 150, and 151-emerged as strong predictors of antigenic relationships. Our results demonstrate the utility of integrating routinely generated genomic data with machine learning to inform vaccine candidate selection and anticipate immune interactions among circulating FMDV strains. This approach offers a practical tool for accelerating vaccine decision-making and can be adapted to other FMDV serotypes. The latest version of the r1 predictive model is available for access via a Shiny dashboard (https://dmakau.shinyapps.io/PredImmune-FMD/).

Foot-and-Mouth Disease Virus↗

PubMed2024

Massively parallel approaches for characterizing noncoding functional variation in human evolution.

The genetic differences underlying unique phenotypes in humans compared to our closest primate relatives have long remained a mystery. Similarly, the genetic basis of adaptations between human groups during our expansion across the globe is poorly characterized. Uncovering the downstream phenotypic consequences of these genetic variants has been difficult, as a substantial portion lies in noncoding regions, such as cis-regulatory elements (CREs). Here, we review recent high-throughput approaches to measure the functions of CREs and the impact of variation within them. CRISPR screens can directly perturb CREs in the genome to understand downstream impacts on gene expression and phenotypes, while massively parallel reporter assays can decipher the regulatory impact of sequence variants. Machine learning has begun to be able to predict regulatory function from sequence alone, further scaling our ability to characterize genome function. Applying these tools across diverse phenotypes, model systems, and ancestries is beginning to revolutionize our understanding of noncoding variation underlying human evolution.

Humans↗

PubMed2025

Generative AI Models in Time-Varying Biomedical Data: Scoping Review.

BACKGROUND: Trajectory modeling is a long-standing challenge in the application of computational methods to health care. In the age of big data, traditional statistical and machine learning methods do not achieve satisfactory results as they often fail to capture the complex underlying distributions of multimodal health data and long-term dependencies throughout medical histories. Recent advances in generative artificial intelligence (AI) have provided powerful tools to represent complex distributions and patterns with minimal underlying assumptions, with major impact in fields such as finance and environmental sciences, prompting researchers to apply these methods for disease modeling in health care. OBJECTIVE: While AI methods have proven powerful, their application in clinical practice remains limited due to their highly complex nature. The proliferation of AI algorithms also poses a significant challenge for nondevelopers to track and incorporate these advances into clinical research and application. In this paper, we introduce basic concepts in generative AI and discuss current algorithms and how they can be applied to health care for practitioners with little background in computer science. METHODS: We surveyed peer-reviewed papers on generative AI models with specific applications to time-series health data. Our search included single- and multimodal generative AI models that operated over structured and unstructured data, physiological waveforms, medical imaging, and multi-omics data. We introduce current generative AI methods, review their applications, and discuss their limitations and future directions in each data modality. RESULTS: We followed the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines and reviewed 155 articles on generative AI applications to time-series health care data across modalities. Furthermore, we offer a systematic framework for clinicians to easily identify suitable AI methods for their data and task at hand. CONCLUSIONS: We reviewed and critiqued existing applications of generative AI to time-series health data with the aim of bridging the gap between computational methods and clinical application. We also identified the shortcomings of existing approaches and highlighted recent advances in generative AI that represent promising directions for health care modeling.

Artificial Intelligence↗

PubMed1996

A generalized hidden Markov model for the recognition of human genes in DNA.

We present a statistical model of genes in DNA. A Generalized Hidden Markov Model (GHMM) provides the framework for describing the grammar of a legal parse of a DNA sequence (Stormo & Haussler 1994). Probabilities are assigned to transitions between states in the GHMM and to the generation of each nucleotide base given a particular state. Machine learning techniques are applied to optimize these probabilities using a standardized training set. Given a new candidate sequence, the best parse is deduced from the model using a dynamic programming algorithm to identify the path through the model with maximum probability. The GHMM is flexible and modular, so new sensors and additional states can be inserted easily. In addition, it provides simple solutions for integrating cardinality constraints, reading frame constraints, "indels", and homology searching. The description and results of an implementation of such a gene-finding model, called Genie, is presented. The exon sensor is a codon frequency model conditioned on windowed nucleotide frequency and the preceding codon. Two neural networks are used, as in (Brunak, Engelbrecht, & Knudsen 1991), for splice site prediction. We show that this simple model performs quite well. For a cross-validated standard test set of 304 genes [ftp:@www-hgc.lbl.gov/pub/genesets] in human DNA, our gene-finding system identified up to 85% of protein-coding bases correctly with a specificity of 80%. 58% of exons were exactly identified with a specificity of 51%. Genie is shown to perform favorably compared with several other gene-finding systems.

Chromosomes, Human↗

PubMed2026

Multimodal artificial intelligence and machine learning in oncology: from data integration to precision cancer care.

Cancer remains a major global health burden, with approximately 20 million new cases and 9.7 million cancer-related deaths reported globally in 2022. While advances in radiological imaging, molecular profiling, and clinical data have enhanced the interpretation of disease progression, the availability of multiple such modalities still does not meet the needs of a large patient population. This narrative review focuses on the role of multimodal artificial intelligence and machine learning in bridging the gap in interpreting heterogeneous modalities to improve risk prediction, prognostic assessment, and treatment decision-making in precision oncology. Multimodal frameworks such as Pathomic Fusion illustrate how complementary histopathological and genomic information can be integrated for cancer diagnosis and prognostic modeling. Multimodal models have demonstrated potential in virtual biopsy, cancer screening, prognostic prediction, radiotherapy planning, intraoperative guidance, and clinical-trial design using digital twins and synthetic control arms. The major limitations of incorporating multimodal artificial intelligence and machine learning in oncology include data heterogeneity, demographic or institutional biases, and reproducibility challenges that hinder translation. Accordingly, appropriate data-governance strategies, fairness audits, and privacy-preserving approaches such as federated learning should be considered where appropriate. Future progress will depend on the development of standardized benchmarking datasets, robust external validation, seamless integration with electronic health records and picture archiving and communication systems, and the implementation of explainable, secure, and clinically validated multimodal artificial intelligence frameworks that support precision oncology in routine clinical practice.

deep learning↗

PubMed2026

Future promise, current clinical ambiguity: a systematic review of machine learning algorithm outputs predicting risk of cardiovascular disease.

OBJECTIVE: To examine whether the outputs of machine learning algorithms designed to predict risk of cardiovascular disease (CVD) address known deficiencies of the Framingham Risk Score (FRS) and improve risk estimates. METHODS: For this critical review, Medline, Embase and IEEE were searched from inception to 1 January 2025. Included were studies describing machine learning algorithms designed to specifically compare output of cardiovascular risk assessment with the FRS. Commentaries, letters, unpublished work or non-peer-reviewed papers were excluded.Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, two reviewers screened titles and abstracts independently, then populated a purpose-built data extraction form. A subsequent qualitative thematic analysis focused on algorithms' strengths, added value, potential harms, unintended consequences and equity implications.The main outcome assessed was whether, among healthy adults, the algorithm improved CVD risk prediction relative to the FRS. RESULTS: Of 707 studies retrieved, 29 met inclusion criteria. 23 reported improved predictive ability relative to the FRS. Most datasets and/or medical records used included sociodemographic predictors of CVD not included among FRS inputs. Some added costly diagnostic tests like CT angiography to FRS screening indicators. When they were defined, inputs and outcomes such as hypertension or myocardial infarction did not always adhere to FRS values. Statistical significance was generally taken as a proxy for clinical significance. Some algorithms overestimated the number at risk compared with the FRS without discussing whether that larger proportion might be at risk of overdiagnosis rather than CVD, while a few decreased the proportion found to be at risk. CONCLUSIONS: Use of artificial intelligence to improve accuracy of risk assessment for CVD demonstrates the technological capacity to merge known sociodemographic predictors with biologic variables and examine non-linear interactions among these. Still needed to achieve patient benefit is clinical insight, adherence to screening principles and cost-benefit assessment of inputs selected.

Humans↗

PubMed2025

Multi‑omics identification of a novel signature for serous ovarian carcinoma in the context of 3P medicine and based on twelve programmed cell death patterns: a multi-cohort machine learning study.

BACKGROUND: Predictive, preventive, and personalized medicine (PPPM/3PM) is a strategy aimed at improving the prognosis of cancer, and programmed cell death (PCD) is increasingly recognized as a potential target in cancer therapy and prognosis. However, a PCD-based predictive model for serous ovarian carcinoma (SOC) is lacking. In the present study, we aimed to establish a cell death index (CDI)-based model using PCD-related genes. METHODS: We included 1254 genes from 12 PCD patterns in our analysis. Differentially expressed genes (DEGs) from the Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) were screened. Subsequently, 14 PCD-related genes were included in the PCD-gene-based CDI model. Genomics, single-cell transcriptomes, bulk transcriptomes, spatial transcriptomes, and clinical information from TCGA-OV, GSE26193, GSE63885, and GSE140082 were collected and analyzed to verify the prediction model. RESULTS: The CDI was recognized as an independent prognostic risk factor for patients with SOC. Patients with SOC and a high CDI had lower survival rates and poorer prognoses than those with a low CDI. Specific clinical parameters and the CDI were combined to establish a nomogram that accurately assessed patient survival. We used the PCD-genes model to observe differences between high and low CDI groups. The results showed that patients with SOC and a high CDI showed immunosuppression and hardly benefited from immunotherapy; therefore, trametinib_1372 and BMS-754807 may be potential therapeutic agents for these patients. CONCLUSIONS: The CDI-based model, which was established using 14 PCD-related genes, accurately predicted the tumor microenvironment, immunotherapy response, and drug sensitivity of patients with SOC. Thus this model may help improve the diagnostic and therapeutic efficacy of PPPM.

Humans↗

PubMed2025

HINN: Hierarchical Input Neural Network identifies multi-omics biomarker for cognitive decline.

Understanding complex diseases requires models that can integrate diverse layers of biological data while yielding insights that are biologically interpretable. Although multi-omics integration with machine learning (ML) has advanced disease prediction and biomarker discovery, most existing approaches overlook the hierarchical and regulatory relationships that connect these molecular layers. Here, we present the Hierarchical Input Neural Network (HINN), a deep learning framework that incorporates known cross-omics relationships directly into its architecture, capturing the flow of information from genomics to epigenomics, transcriptomics, and downstream biological processes. By embedding these relationships, HINN improves both predictive performance and biological interpretability. We applied HINN to blood-derived multi-omics data from individuals with Alzheimer's disease or mild cognitive impairment to predict cognitive scores from standardized assessments. HINN outperformed both baseline and state-of-the-art models and pinpointed multi-omics biomarkers-including SNPs and promoter-region CpG sites in ATP6V1C1 and RCHY1 -that were significantly correlated with plasma p-Tau181 levels. These features map to biologically relevant processes with potential implications for cognitive decline. Our findings demonstrate how combining deep learning with biological knowledge can uncover interpretable, blood-based biomarkers for cognitive decline due to complex diseases such as Alzheimer's. All code and data are openly available at https://github.com/bozdaglab/HINN.

Alzheimer’s disease↗

PubMed2026

SeqQC-former: A sequence-quality fusion framework for QC-aware review prioritization of candidate somatic SNVs in cancer genomics.

The accurate prioritization of candidate somatic single-nucleotide variants (SNVs) remains a challenge due to the substantial variability in sequencing quality across genomic loci. SeqQC-Former is a sequence-quality fusion framework that integrates the local nucleotide context with read-level quality-control (QC) covariates derived from matched tumor-normal sequencing data. This integration generates QC-aware prioritization scores for the downstream review of candidate variants. Unlike conventional variant callers, SeqQC-Former is designed not to infer biological truth but to support post-calling review and prioritization under heterogeneous sequencing conditions. The framework was trained and evaluated on a SEQC2-derived dataset comprising 89,447 candidate loci, including 1378 positive and 88,069 negative loci. In chromosome-held-out validation, which aims to reduce potential genomic-position leakage, SeqQC-Former demonstrated strong discrimination (AUROC = 0.9479; AUPRC = 0.9448), indicating good generalization to previously unseen chromosomes. Given that the SEQC2-derived labels contain QC-associated information; these results should be interpreted as an evaluation of QC-aware prioritization capability rather than an independent validation of biological variant correctness. Ablation analyses revealed that structured QC covariates provided the dominant predictive signal under the current SEQC2-derived labeling regime. SeqQC-Former achieved a significantly higher AUROC than classical machine-learning baselines, as determined by DeLong's test (p < 0.01). Application to 53,164 glioblastoma variants demonstrated that external predictions were sensitive to QC scaling and threshold selection, underscoring that model outputs should be interpreted as QC-dependent prioritization scores rather than calibrated probabilities or definitive biological classifications. Overall, SeqQC-Former offers a reproducible post-calling QC-aware prioritization framework for large-scale somatic SNV review and underscores the importance of explicitly modeling sequencing-quality information when interpreting structured cancer genomics datasets.

Humans↗

PubMed2026

Fine-grained structural classification of biosynthetic gene cluster-encoded products.

MOTIVATION: Biosynthetic gene clusters (BGCs) are responsible the biosynthesis of many natural products, including a multitude of effective therapeutics and their precursors. Advances in genomic data collection as well as computational techniques have made it possible to identify BGCs at scale. However, accurately determining the types of BGC-encoded products from genomic content remains elusive. RESULTS: Here, we introduce BGC annotation tool (BGCat), a machine learning method for fine-grained structural classification of BGC-encoded products, leveraging the NPClassifier natural product nomenclature. Our method leverages a pre-trained protein language model for creating meaningful gene representations and a deep neural network for class label prediction. We show the method outperforms state-of-the-art approaches in coarse-grained product classification and is effective for detailed classification. We implement a clustering-based augmentation strategy for BGC-product relationships, addressing a crucial gap in the available datasets. We then introduce the concept of product class profiles of gene cluster families (GCFs), associating each GCF with a probabilistic distribution of product types and offering a new perspective on GCF functions. Lastly, we use BGCat to provide new product class labels for over 100k BGCs in antiSMASH DB that presently have minimal information about their products. AVAILABILITY AND IMPLEMENTATION: The source code and trained model weights are freely available at https://github.com/HassounLab/BGCat.

Multigene Family↗

PubMed2025

Assessing data size requirements for training generalizable sequence-based TCR specificity models via pan-allelic MHC-I point-mutation ligandome evaluation.

Rapid identification of T cell receptors (TCRs) that specifically bind patient-unique neoepitopes is a critical challenge for personalized TCR-based therapies in oncology. Due to enormous diversity of both TCR and neoepitope repertoires, a machine learning predictor of TCR-pMHC specificity for personalized therapy must generalize to TCRs and epitopes not seen in the training data. We estimate the necessary size of such training data. We first confirm that published models fail to generalize beyond a single-residue dissimilarity to the epitope training set distribution. We then impute the point-mutation ligandome across the 34 most prevalent human MHC alleles and represent it as a graph based on our established dissimilarity cutoff. By finding the dominating set of this graph, we estimate that between one and 100 million epitopes are required to train a generalizable sequence-based TCR specificity prediction model-1000 times the size of current public data.

Humans↗

PubMed2025

Genomic Language Model for Predicting Enhancers and Their Allele-Specific Activity in the Human Genome.

Predicting and deciphering the regulatory logic of enhancers is a challenging problem, due to the intricate sequence features and lack of consistent genetic or epigenetic signatures that can accurately discriminate enhancers from other genomic regions. Recent machine-learning based methods have spotlighted the importance of extracting nucleotide composition of enhancers but failed to learn the sequence context and perform suboptimally. Motivated by advances in genomic language models, we developed DNABERT-Enhancer, a novel enhancer prediction method, by applying DNABERT pre-trained language model on the human genome. We trained two different models, using large collection of enhancers curated from the ENCODE registry of candidate cis-Regulatory Elements. The best fine-tuned model achieved 88.05% accuracy with Matthews correlation coefficient of 76% on independent set aside data. Further, we present the analysis of the predicted enhancers for all chromosomes of the human genome by comparing with the enhancer regions reported in publicly available databases. Finally, we applied DNABERT-Enhancer along with other DNABERT based regulatory genomic region prediction models to predict candidate SNPs with allele-specific enhancer and transcription factor binding activity. The genome-wide enhancer annotations and candidate loss-of-function genetic variants predicted by DNABERT-Enhancer provide valuable resources for genome interpretation in functional and clinical genomics studies.

Journal Article↗

PubMed1998

Contextual quick-learning and generalization by humans and machines.

In a previous study (1994 Network: Comput. Neural Syst. 5 203-27) we compared human quick-learning and generalization (quick modelling) with that of neural nets (feedforward architectures), symbolic algorithms (decision tree procedures), and pattern classifiers (truth-set descriptors). Those studies raised the question of the role of context in the nature and rapidity of human learning. Here we address that issue in the setting of the same basic experiment (Quinlan classification problem) used for the previous studies. A major implication of our findings is that humans overwhelmingly seek, create, or imagine context in order to provide meaning when presented with abstract or apparently incomplete or contradictory or otherwise untenable situations.

Adolescent↗

PubMed2022

Integration of Infant Metabolite, Genetic, and Islet Autoimmunity Signatures to Predict Type 1 Diabetes by Age 6 Years.

CONTEXT: Biomarkers that can accurately predict risk of type 1 diabetes (T1D) in genetically predisposed children can facilitate interventions to delay or prevent the disease. OBJECTIVE: This work aimed to determine if a combination of genetic, immunologic, and metabolic features, measured at infancy, can be used to predict the likelihood that a child will develop T1D by age 6 years. METHODS: Newborns with human leukocyte antigen (HLA) typing were enrolled in the prospective birth cohort of The Environmental Determinants of Diabetes in the Young (TEDDY). TEDDY ascertained children in Finland, Germany, Sweden, and the United States. TEDDY children were either from the general population or from families with T1D with an HLA genotype associated with T1D specific to TEDDY eligibility criteria. From the TEDDY cohort there were 702 children will all data sources measured at ages 3, 6, and 9 months, 11.4% of whom progressed to T1D by age 6 years. The main outcome measure was a diagnosis of T1D as diagnosed by American Diabetes Association criteria. RESULTS: Machine learning-based feature selection yielded classifiers based on disparate demographic, immunologic, genetic, and metabolite features. The accuracy of the model using all available data evaluated by the area under a receiver operating characteristic curve is 0.84. Reducing to only 3- and 9-month measurements did not reduce the area under the curve significantly. Metabolomics had the largest value when evaluating the accuracy at a low false-positive rate. CONCLUSION: The metabolite features identified as important for progression to T1D by age 6 years point to altered sugar metabolism in infancy. Integrating this information with classic risk factors improves prediction of the progression to T1D in early childhood.

Autoantibodies↗

PubMed2026

How advances in machine learning drive early detection and risk prediction of early-onset colorectal cancer.

Early-onset colorectal cancer (EOCRC), defined as colorectal cancer diagnosed before age 50, is rising across high- and middle-income settings whilst organised screening stays anchored to older age thresholds. Blood-based liquid biopsy, combined with machine learning, is the most plausible route to early detection in this group because it does not depend on bowel preparation, endoscopy capacity, or adherence to stool-based testing. The gap is structural: incidence climbs fastest in the population below the age at which any guideline-endorsed modality is offered. The analytical challenge is that early-stage tumour-derived signals in plasma are low in abundance and distributed across heterogeneous molecular layers: circulating tumour DNA mutations, aberrant methylation, cfDNA fragmentomics, and small non-coding RNA. Machine learning converts these into a single calibrated probability. This review examines where artificial intelligence (AI)-driven liquid biopsy genuinely adds diagnostic value in EOCRC, distinguishes components in which learned models are decorative from those in which they are mechanistically necessary, and identifies the validation deficit separating research cohorts from deployable clinical tools. It summarises the first-generation tools used clinically for early detection and post-treatment monitoring, then considers analytes from exosome-bound microRNAs to long-read whole-genome sequencing of circulating plasma DNA, which reads cytosine modification natively, resolves methylation and fragmentation on single molecules, and characterises structural events short reads cannot anchor. Any analyte can feed a learned model, but more diverse input yields better discrimination. The central argument is that approved, guideline-included blood tests were validated in populations aged 45 and above, and their performance in younger patients cannot be assumed.

cfDNA fragmentomics↗

PubMed2023

Human mutations in high-confidence Tourette disorder genes affect sensorimotor behavior, reward learning, and striatal dopamine in mice.

UNLABELLED: Tourette disorder (TD) is poorly understood, despite affecting 1/160 children. A lack of animal models possessing construct, face, and predictive validity hinders progress in the field. We used CRISPR/Cas9 genome editing to generate mice with mutations orthologous to human de novo variants in two high-confidence Tourette genes, CELSR3 and WWC1 . Mice with human mutations in Celsr3 and Wwc1 exhibit cognitive and/or sensorimotor behavioral phenotypes consistent with TD. Sensorimotor gating deficits, as measured by acoustic prepulse inhibition, occur in both male and female Celsr3 TD models. Wwc1 mice show reduced prepulse inhibition only in females. Repetitive motor behaviors, common to Celsr3 mice and more pronounced in females, include vertical rearing and grooming. Sensorimotor gating deficits and rearing are attenuated by aripiprazole, a partial agonist at dopamine type II receptors. Unsupervised machine learning reveals numerous changes to spontaneous motor behavior and less predictable patterns of movement. Continuous fixed-ratio reinforcement shows Celsr3 TD mice have enhanced motor responding and reward learning. Electrically evoked striatal dopamine release, tested in one model, is greater. Brain development is otherwise grossly normal without signs of striatal interneuron loss. Altogether, mice expressing human mutations in high-confidence TD genes exhibit face and predictive validity. Reduced prepulse inhibition and repetitive motor behaviors are core behavioral phenotypes and are responsive to aripiprazole. Enhanced reward learning and motor responding occurs alongside greater evoked dopamine release. Phenotypes can also vary by sex and show stronger affection in females, an unexpected finding considering males are more frequently affected in TD. SIGNIFICANCE STATEMENT: We generated mouse models that express mutations in high-confidence genes linked to Tourette disorder (TD). These models show sensorimotor and cognitive behavioral phenotypes resembling TD-like behaviors. Sensorimotor gating deficits and repetitive motor behaviors are attenuated by drugs that act on dopamine. Reward learning and striatal dopamine is enhanced. Brain development is grossly normal, including cortical layering and patterning of major axon tracts. Further, no signs of striatal interneuron loss are detected. Interestingly, behavioral phenotypes in affected females can be more pronounced than in males, despite male sex bias in the diagnosis of TD. These novel mouse models with construct, face, and predictive validity provide a new resource to study neural substrates that cause tics and related behavioral phenotypes in TD.

Preprint↗

PubMed2025

STRUMP-I: Structure-based machine learning approach to pMHC-I binding prediction using force field energy features.

The adaptive immune system monitors cellular integrity by recognizing short peptides from intracellular proteins presented on Major Histocompatibility Complex class I (MHC-I) molecules, collectively termed peptide-MHC complexes (pMHC), enabling detection of foreign or mutated proteins. With the rising importance of immunotherapies targeting neoantigens in cancers, the ability to accurately predict which peptides will bind to the diverse population of MHC alleles is critically important. Current computational methods for pMHC-I prediction fall broadly into sequence-based methods, which rely heavily on large training datasets, and structure-based methods that leverage structural modeling and energetics of pMHC binding. While sequence-based methods have been popularly used, their performance is dependent on the size and quality of training data. On the other hands, while structure-based approaches can generalize better across diverse MHC alleles, they traditionally depend on identifying a single global minimum energy conformation, an assumption that often fails due to the inherent binding promiscuity of MHC-I molecules. To address these limitations, we developed a STRUMP-I (STRUcture-based pMHC Prediction (for class I)), a novel pMHC binding prediction tool that directly leverages a broad set of force-field-derived energy terms as machine-learning features. STRUMP-I achieves performance comparable to state-of-the-art sequence-based models while significantly outperforming them on MHC alleles with limited representation in training data. Furthermore, STRUMP-I demonstrates strong synergy when integrated with sequence-based methods, notably enhancing prediction precision. The robustness and generalizability of STRUMP-I were confirmed by evaluating its predictive performance on independent, previously unseen datasets, including an experimentally validated cancer neoantigen dataset. This combined approach advances our capability to reliably identify clinically relevant neoantigen targets. The source code and trained models are available at https://github.com/yoonjoolab/STRUMP-I.

energy optimization↗