Biology subjects

Lin, G. N.

Publications and source records attributed to Lin, G. N..

12 recordsLinked to original sources

bioRxiv2026

BaiZe: A Multi-View Dynamic Framework for Simulating and Interpreting Cellular Responses Across Perturbation Contexts

Accurately predicting how cells respond to perturbations is important for understanding cellular regulation and prioritizing experimental interventions, yet existing models are often designed for specific perturbation types or biological contexts. Here we present BaiZe, a multi-view conditional state-transition framework that predicts the post-perturbation transcriptome from a control-state transcriptome together with genetic, chemical, temporal and optional chromatin-accessibility information. BaiZe models perturbation responses as context-dependent transitions between cellular states. BaiZe supports prediction across held-out cell states and genetic perturbations, unseen multi-gene combinations, chemical structures and doses, temporal stages and species contexts. Benchmarking across diverse perturbation settings demonstrates that BaiZe effectively recovers major transcriptional response programs under previously unseen conditions. Incorporating matched ATAC-seq context further improves selected state-transition predictions and enables model-based attribution of chromatin regions to response-associated genes and pathways. BaiZe also supports few-shot transfer of perturbation responses from human to mouse cellular systems and connects predicted transcriptomes to candidate morphology projections. To facilitate interpretation and use, BaiZe-Agent organizes response genes, pathways, chromatin evidence, cross-species predictions and projected phenotypes into traceable, queryable perturbation records. Together, BaiZe provides a broadly applicable framework for predicting and interpreting context-dependent cellular responses and for prioritizing hypotheses across diverse perturbation settings.

bioinformatics↗

bioRxiv2025

An Interactive AI Agent for Adaptive Modeling of RNA Region-Ligand Interactions via LLM-Generated Machine Learning Workflows

Precise modeling of RNA-ligand interactions is essential for understanding RNA functionality and designing RNA-targeted therapeutics. Current computational approaches largely focus on predicting discrete binding sites, limiting their applicability to complex RNA regions that may harbor multiple or diffuse ligand binding motifs. Here, we present RLAgent, an interactive agent framework designed to predict ligand interactions at the RNA region level, enabling higher-resolution and more flexible modeling than conventional site-centric approaches. RLAgent reframes the RNA-ligand prediction workflow as a dialogue-driven process. Through a natural language interface, users can interactively configure modeling preferences without writing code. A locally hosted large language model (LLM) acts as the core orchestration agent, automating all key components of the modeling pipeline, including data validation, feature encoding, model training, evaluation, and visualization. This agent-based design lowers technical barriers and enhances reproducibility, making RNA-ligand prediction more accessible for both computational and experimental researchers.

bioinformatics↗

bioRxiv2025

Memo-Patho: Bridging Local-Global Transmembrane Protein Contexts with Contrastive Pretraining for Alignment-Free Pathogenicity Prediction

Understanding the pathogenic impact of protein mutations remains a fundamental challenge in genomic medicine, particularly for transmembrane proteins (TMPs), a functionally critical yet structurally under-annotated class that includes numerous drug targets. Current variant effect predictors often fall short in this domain due to computational burdens, reliance on multiple sequence alignments (MSAs), and limited ability to generalize to TMP-specific constraints. Here we present Memo-Patho, an alignment-free deep learning framework specifically optimized for TMP mutation pathogenicity prediction. Memo-Patho integrates global and local representations from protein language models (PLMs) and predicted structural features. Furthermore, it introduces a contrastive pre-training strategy that learns discriminative features by comparing benign and pathogenic mutations within the same protein backbone. This design enables the model to capture functional disruptions in challenging contexts without structural or evolutionary alignments. Memo-Patho outperforms state-of-the-art predictors on two types of TMP benchmark datasets (accuracy up to 0.93) and unseen-protein generalization tasks, showing strong agreement with evolutionary conservation profiles. Its predictions are further validated on a manually curated KCNQ1 dataset of 55 ion channel variants, achieving an accuracy of 0.84 and outperforming existing tools. The model also supports high-throughput scanning, efficiently analyzing over 15000 variants in under 80 minutes. Our results establish Memo-Patho as a fast, generalizable, and biologically grounded approach for variant effect prediction in TMPs. Its alignment-free design, contrastive learning paradigm, and clinical robustness mark a step forward towards scalable interpretation of TMP variation for both basic research and precision medicine applications.

bioinformatics↗

bioRxiv2025

Benchmarking the coding strategies of non-coding mutations on sequence-based downstream tasks with machine learning

Non-coding single nucleotide polymorphisms (SNPs) are key modulators of gene regulation and have been implicated in diverse complex traits and diseases. With the growing demand for accurate functional interpretation of non-coding variants, the choice of encoding strategies becomes critical in downstream predictive modeling. Despite recent advances, a systematic evaluation of encoding approaches tailored for non-coding SNPs remains lacking. To address this gap, we present a comprehensive benchmark that evaluates six representative encoding strategies, including categorical, semantic, and functional embeddings, across three quantitative trait loci (QTL)-related prediction tasks. The study encompasses nine machine learning and deep learning models and incorporates experimental controls and repeated trials to ensure robustness and reproducibility. We assess each strategy along multiple dimensions, such as interpretability, representation abundance, and computational efficiency. Rather than ranking individual methods, our analysis emphasizes the interaction between encoding strategies, model types, and preprocessing protocols, and highlights their collective influence on predictive performance. This work establishes a standardized framework for evaluating non-coding SNP representations and offers actionable guidance for selecting and optimizing prediction pipelines in regulatory genomics.

bioinformatics↗

bioRxiv2024

Predicting the effect of non-coding mutations on single-cell DNA methylation using deep learning

Predicting the effects of non-coding mutations on DNA methylation is crucial for advancing our understanding of gene expression, epigenetic inheritance, and its role in disease mechanisms. Current methods lack the capability to predict the impact of non-coding mutations on DNA methylation at single-cell resolution and long range, while remain challenges in tracking SNP influences throughout disease progression. Here, we introduce Methven, a deep learning-based framework designed to predict the effects of non-coding mutations on DNA methylation at single-cell resolution, to overcome the challenges. Methven integrates DNA sequences and ATAC-seq data, employing a divide-and-conquer approach to handle varying scales of SNP-CpG interactions. By leveraging a pretrained DNA language model, Methven accurately predicts both the direction and magnitude of methylation changes across a 100kbp range with a lightweight architecture. The evaluation results demonstrate the superior performance of Methven in prioritizing functional non-coding mutation, model interpretability, and its potential for revealing personalized mutation-disease associations.

bioinformatics↗

bioRxiv2023

EMO: Predicting Non-coding Mutation-induced Up- and Down-regulation of Risk Gene Expression using Deep Learning

The challenge of understanding how alterations in non-coding DNA regulate gene expression is substantial, with far-reaching consequences for the advancement of human genetics and disease research. Accurately predicting the up- and down-regulation of gene expression quantitative trait loci (eQTLs) offers a potential avenue to accelerate the identification of associations between non-coding variants and phenotypic traits. However, current methods for predicting the impact of non-coding mutations on gene expression changes fail to predict the sign of eQTLs accurately. Additionally, the requirement for tissue-specific training models within these methods restricts their applicability, especially when extending predictive abilities to single-cell resolution. In this study, we present EMO, an innovative transformer-based pre-trained method, designed to predict the up- and down-regulation of gene expression caused by single non-coding mutations using DNA sequences and ATAC-seq data. EMO extends the effective prediction range up to 1Mbp between the non-coding mutation and the transcription start site (TSS) of the target gene. It demonstrates competitive prediction performance across various variant TSS distances and surpasses the state-of-the-art structure. To assess its effectiveness, EMO was fine-tuned using eQTLs from two brain tissues for external validation. We also evaluated EMOs transferability to single-cell resolution by fine-tuning it on eQTLs from six types of immune cells, achieving satisfactory results in each cell type (AUC > 0.860). Furthermore, EMO displayed promising potential in analyzing disease-associated eQTLs.

bioinformatics↗

bioRxiv2023

Haplotype Function Score improves biological interpretation and cross-ancestry polygenic prediction of human complex traits

We propose a new framework for human genetic association studies: at each locus, a deep learning model (in this study, Sei) is used to calculate the functional genomic activity score for two haplotypes per individual. This score, defined as the Haplotype Function Score (HFS), replaces the original genotype in association studies. Applying the HFS framework to 14 complex traits in the UK Biobank, we identified 3,619 independent HFS-trait associations with a significance of p<5x10-8. Fine-mapping revealed 2,699 causal associations, corresponding to a median increase of 63 causal findings per trait compared with SNP-based analysis. HFS-based enrichment analysis uncovered 727 pathway-trait associations and 153 tissue-trait associations with strong biological interpretability, including "circadian pathway-chronotype" and "arachidonic acid-intelligence". Lastly, we applied LASSO regression to integrate HFS prediction score with SNP-based polygenic risk scores, which showed an improvement of 16.1% to 39.8% in cross-ancestry polygenic prediction. We concluded that HFS is a promising strategy for understanding the genetic basis of human complex traits.

genetics↗

bioRxiv2023

A small number of human lineage mutations regulated RNA-protein binding of conserved genes and promoted human evolution

Whether human lineage mutations (HLM) have contributed to human evolution via post-transcriptional modification remained unknown. We applied deep learning models Seqweaver to predict how HLM impacts RNA-binding protein affinity (RBP). At the threshold of the top 1% of human common variation, only 0.27% of HLM had large impacts on RBP. These HLMs enriched in a set of conserved genes that are highly expressed in adult excitatory neurons and prenatal Purkinje neurons, and involved in synapse organization and the GTPase pathway. These genes also carried excess damaging coding mutations that caused neurodevelopmental disorders, ataxia, and Schizophrenia. Among these genes, NTRK2 and ITPR1 had the most aggregated evidence of functional importance, pointing to an essential role in cognition and bipedalism. We concluded that a very small number of human-specific mutations have contributed to human speciation via impacts on post-transcriptional modification of critical brain-related genes. Author summaryPost-transcriptional modification has important functions in biological systems, but its role in evolution has long been a mystery due to the difficulty in predicting how a mutation could impact this process. Here we applied the newly developed deep learning model Seqweaver on human lineage mutations (HLM) to predict their effect on RNA-protein binding affinity profile (RBP). We found that only a very small number HLM are influential to RBP. Influential HLMs enriched in a set of conserved genes that are enriched in cells and functions critical for cognition and bipedal walking, and severe mutations on these genes cause the disruption of cognition (neurodevelopmental disorders) and bipedalism (ataxia). NTRK2 and ITPR1 had the most aggregated pieces of evidence of functional importance, pointing to an important role in cognition and bipedal walking. Taken together, our result demonstrated that there is a small number of human lineage mutations that modified the post-transcriptional modifications of conserved neuronal genes, which contributed to human speciation.

evolutionary biology↗

bioRxiv2020

Isoform transcriptome of developing human brain provides new insights into autism risk variants

Alternative splicing plays important role in brain development, however its global contribution to human neurodevelopmental diseases (NDD) has not been fully investigated. Here, we examined the relationships between full-length splicing isoforms expression in the brain and de novo loss-of-function mutations identified in the patients with NDDs. We analyzed the full-length isoform transcriptome of the developing human brain and observed differentially expressed isoforms and isoform co-expression modules undetectable by gene-level analyses. These isoforms were enriched in loss-of-function mutations and microexons, co-expressed with a unique set of partners, and had higher prenatal expression. We experimentally tested the impact of splice site mutations in five NDD risk genes, including SCN2A, DYRK1A and BTRC, and demonstrated exon skipping. Furthermore, our results suggest that the splice site mutation in BTRC reduces translational efficiency, likely impacting Wnt signaling through impaired degradation of {beta}-catenin. We propose that functional effect of mutations associated with human diseases should be investigated at the isoform-rather than the gene-level resolution. HighlightsO_LIDifferential isoform expression analysis of the human brain transcriptome reveals neurodevelopmental processes and pathways undetectable by differential gene expression analyses. C_LIO_LISplicing isoforms impacted by neurodevelopmental disease (NDD) risk mutations exhibit higher prenatal expression, are enriched in microexons and are involved in neuronal-related functions. C_LIO_LIIsoform co-expression network analysis identifies modules with splicing and synaptic functions that are enriched in NDD mutations. C_LIO_LISplice site mutations impacting NDD risk genes cause exon skipping and produce novel isoforms with altered biological properties. C_LIO_LIFunctional impact of mutations should be investigated at the full-length isoform-level rather than the gene-level resolution C_LI

neuroscience↗

bioRxiv2020

K-means Based Unsupervised Feature Selection to Prioritize Biomarkers of Different Disease Clinical Phases

Huntingtons disease is caused by a single gene mutation, which is potentially a good model for development of biomarkers corresponding to different disease phase and clinical phenotypes. Hypothesis-driven and omics discovery approaches have not yet identified effective candidate biomarkers in HD. So, it is urgent to develop engagement and disease-phase specific biomarkers. The advanced sequencing technology makes it possible to develop data-driven methods for biomarkers discovery. Therefore, in this study, we designed k-means based unsupervised feature selection (KFS) method to prioritize biomarkers of different disease clinical phases. KFS first conducts k-means clustering on the samples with gene expression data, then it conducts feature selection based on the feature selection matrix to prioritize biomarkers of different samples. By conducting alternative iteration of clustering and feature selection to screen key genes which corresponding to the complex clinical phenotypes of different disease phases. Further gene ontology and enrichment analysis highlight potential molecular mechanisms of HD. Our experimental analyses have uncovered new disease-related genes and disease-associated pathways, which in turn have provided insight into the molecular mechanisms during the disease progression.

bioinformatics↗

bioRxiv2019

PsyMuKB: A De Novo Variant Knowledge Base Integrating Transcriptional and Translational Information to Identify Isoform-specific Mutations in Developmental Disorders

De novo variants (DNVs) are one of the most significant contributors to severe early-onset genetic disorders such as autism spectrum disorder, intellectual disability, and other developmental and neuropsychiatric (DNP) disorders. Currently, a plethora of DNVs have being identified through the use of next-generation sequencing and much effort has been made to understand their impact at the gene level; however, there has been little exploration of the impact at the isoform level. The brain contains a high level of alternative splicing and regulation, and exhibits a more divergent splicing program than other tissues; therefore, it is crucial to explore variants at the transcriptional regulation level to better interpret the mechanisms underlying DNP disorders. To facilitate better usage and improve the isoform-level interpretation of variants, we developed the PsyMuKB (NeuroPsychiatric Mutation Knowledge Base), a knowledge base containing a comprehensive, carefully curated list of DNVs with transcriptional and translational annotations to enable identification of isoform-specific mutations. PsyMuKB allows a flexible search of genes or variants and provides both table-based descriptions and associated visualizations, such as expression, transcript genomic structures, protein interactions, and the mutation sites mapped on the protein structures. It also provides an easy-to-use web interface, allowing users to rapidly visualize the locations and characteristics of mutations and the expression patterns of the impacted genes and isoforms. PsyMuKB thus constitutes a valuable resource for identifying tissue-specific de novo mutations for further functional studies of related disorders. PsyMuKB is freely accessible at http://psymukb.net.

bioinformatics↗

bioRxiv2017

MutPred2: inferring the molecular and phenotypic impact of amino acid variants

We introduce MutPred2, a tool that improves the prioritization of pathogenic amino acid substitutions, generates molecular mechanisms potentially causative of disease, and returns interpretable pathogenicity score distributions on individual genomes. While its prioritization performance is state-of-the-art, a novel and distinguishing feature of MutPred2 is the probabilistic modeling of variant impact on specific aspects of protein structure and function that can serve to guide experimental studies of phenotype-altering variants. We demonstrate the utility of MutPred2 in the identification of the structural and functional mutational signatures relevant to Mendelian disorders and the prioritization of de novo mutations associated with complex neurodevelopmental disorders. We then experimentally validate the functional impact of several variants identified in patients with such disorders. We argue that mechanism-driven studies of human inherited diseases have the potential to significantly accelerate the discovery of clinically actionable variants.\n\nAvailability: http://mutpred.mutdb.org/

bioinformatics↗