AI x Bio Discovery

Reengineering the antigen optimization process for superior neoantigen vaccine design

Figure 1

Figure 2

Figure 3

Reengineering the antigen optimization process for superior neoantigen vaccine design [new]
Deep reinforcement learning optimizes neoantigen sequences by improving HLA binding and T cell responses for personalized cancer vaccines.

Figure 1

Figure 2

Figure 3

Figure 4

Toward Efficient Spiking Transformers: Synapse Pruning Meets Synergistic Learning-Based Compensation [new]
Pruning & comp. make light spiking Transformers. Sparsity & low-rank cut weight redun., boosted by neuron model.

A Bayesian approach to model uncertainty in single-cell genomic data

Figure 1

Figure 3

Figure 2

A Bayesian approach to model uncertainty in single-cell genomic data [new]
Models cell identity uncertainty in single-cell data using a Bayesian Gaussian mixture model. Allows probabilistic clustering and improved analysis of cellular transitions.

CellForge: Agentic Design of Virtual Cell Models

Figure 1

Figure 2

Figure 3

CellForge: Agentic Design of Virtual Cell Models [new]
Agentic Design of Virtual Cell Models:
...builds virtual cell models from multi-omics data via collaborative LLM agents. It optimizes model architecture and code generation for perturbation prediction.

Carafe enables high quality in silico spectral library generation for data-independent acquisition proteomics

Figure 1

Figure 2

Figure 3

Carafe enables high quality in silico spectral library generation for data-independent acquisition proteomics [updated]
DL models train on DIA data for experiment-specific, high-quality spectral libraries.

Figure 1

Figure 2

Figure 3

Figure 4

Mechanism of Tethered Agonist Binding to an Adhesion G-Protein-Coupled Receptor [new]
Simulations revealed the dynamic mechanism of a tethered agonist binding to an adhesion GPCR, including conformational changes.

OmniCellAgent: Towards AI Co-Scientists for Scientific Discovery in Precision Medicine

Figure 1

Figure 2

Figure 3

OmniCellAgent: Towards AI Co-Scientists for Scientific Discovery in Precision Medicine [new]
AI agent: scRNA-seq, tools, biomed knowledge. Aids non-experts in precision medicine R&D: mechanism & therapy ID.

Design of Orthogonal Far-Red, Orange and Green Fluorophore-binding Proteins for Multiplex Imaging

Figure 1

Figure 2

Figure 3

Design of Orthogonal Far-Red, Orange and Green Fluorophore-binding Proteins for Multiplex Imaging [new]
Novel proteins bind far-red, orange, green dyes (hi affinity/spec). Multiplex imaging & protein interaction use shown.

Figure 1

Figure 2

Figure 3

Figure 4

A combinatorial mutational map of active non-native protein kinases by deep learning guided sequence design [new]
DL redesign: functional, non-native kinases via combinatorial mutations, explores sequence-function at scale.

Genetic Insights of Image-Based Traits: Analysis Pipeline for AI-based Phenotyping, Combined-GWAS, and Federated Learning with Application to the Human Face

Figure 1

Figure 2

Figure 3

Genetic Insights of Image-Based Traits: Analysis Pipeline for AI-based Phenotyping, Combined-GWAS, and Federated Learning with Application to the Human Face [new]

Scaling down protein language modeling with MSA Pairformer

Figure 1

Figure 2

Figure 3

Scaling down protein language modeling with MSA Pairformer [new]
...Pairformer: memory-efficient MSA, bi-directional updates, better evol. signals, outperforms larger models.

GeneTerrain-GMM Unmasks a Coordinated Neuroinflammatory and Cell Death Network Perturbed by Dasatinib in a Human Neuronal Model of Alzheimer's Disease

Figure 1

Figure 2

Figure 3

GeneTerrain-GMM Unmasks a Coordinated Neuroinflammatory and Cell Death Network Perturbed by Dasatinib in a Human Neuronal Model of Alzheimer's Disease [new]

Learning from B Cell Evolution: Adaptive Multi-Expert Diffusion for AntibodyDesign via Online Optimization

Figure 1

Figure 2

Figure 3

Learning from B Cell Evolution: Adaptive Multi-Expert Diffusion for AntibodyDesign via Online Optimization [new]
Adaptive diffusion model: B cell evolution, multiple physics-based experts, personalized Ab design via online opt.

MicrobioRel: A Set of Datasets for Microbiome Relation Extraction

Figure 2

Figure 3

Table 1

MicrobioRel: A Set of Datasets for Microbiome Relation Extraction [new]
Datasets explore relationships among gut microbiome entities. One dataset is manually annotated, and the other is machine-predicted.

Figure 1

Figure 2

Figure 3

Figure 4

ProteomeLM: A proteome-scale language model allowing fast prediction of protein-protein interactions and gene essentiality across taxa [new]

ET-Pfam: Ensemble transfer learning for protein family prediction

Figure 1

Figure 2

Figure 3

ET-Pfam: Ensemble transfer learning for protein family prediction [new]
Predicts protein families using transfer learning and ensembles of DL classifiers, trained with protein language model representations and weighted voting.

spOT-NMF: Optimal Transport-Based Matrix Factorization for Accurate Deconvolution of Spatial Transcriptomics

Figure 1

Figure 2

Figure 3

spOT-NMF: Optimal Transport-Based Matrix Factorization for Accurate Deconvolution of Spatial Transcriptomics [new]
Ref-free spatial transcriptomics deconvolution: Disentangle mixed gene expr. via optimal transport.

Figure 1

Figure 2

Figure 3

Figure 4

High-content high-resolution microscopy and deep learning assisted analysis reveals host and bacterial heterogeneity during Shigella infection [updated]

Figure 1

Figure 2

Figure 3

Figure 4

Interpretable spatial multi-omics data integration and dimension reduction with SpaMV [new]
Integrates spatial multi-omics, capturing shared/specific info; provides interpretable representations for tasks like clustering.

HPInet: Interpretable prediction of Host-Pathogen protein-protein Interactions using a transformer-based neural network

Figure 1

Figure 2

Figure 3

HPInet: Interpretable prediction of Host-Pathogen protein-protein Interactions using a transformer-based neural network [new]
Predicts effector-host protein interactions via transformer, CNN & GRN; finds key residues from sequence.

Federated Knowledge Retrieval Elevates Large Language Model Performance on Biomedical Benchmarks

Figure 1

Figure 2

Figure 3

Federated Knowledge Retrieval Elevates Large Language Model Performance on Biomedical Benchmarks [new]
LLM accuracy improved by biomed federated knowledge.

Figure 1

Figure 2

Table 1

Table 2

Cost-effective genomic prediction for fertility traits: A comparison of machine learning and conventional models using low-coverage sequencing in Holstein heifers [new]

scDIAGRAM: Detecting Chromatin Compartments from Individual Single-Cell Hi-C Matrix without Imputation or Reference Features

Figure 1

Figure 2

Figure 3

scDIAGRAM: Detecting Chromatin Compartments from Individual Single-Cell Hi-C Matrix without Imputation or Reference Features [new]
A/B compartment ID in scHi-C data via stats & graph analysis. No external data/imputation needed.

Figure 1

Figure 2

Figure 3

Figure 4

A universal model for drug-receptor interactions [new]
Models generalizing drug-target interactions to novel chemistries are lacking; this research presents a theoretical framework to address this.

Heuristic Multi-site Optimization for Protein Sequence Design using Masked Protein Language Models

Figure 1

Figure 3

Figure 5

Heuristic Multi-site Optimization for Protein Sequence Design using Masked Protein Language Models [new]
Masked LMs guide protein opt., mimicking evol. to find functional variants.

CIAdex: Single-Cell FTIR Spectral Fingerprinting for Cell Identity Verification and Aging Quantification in Therapeutic Cell Manufacturing

Figure 1

Figure 2

Figure 3

CIAdex: Single-Cell FTIR Spectral Fingerprinting for Cell Identity Verification and Aging Quantification in Therapeutic Cell Manufacturing [new]

What does it take to learn the rules of RNA base pairing? A lot less than you may think

Figure 1

Figure 2

Figure 3

What does it take to learn the rules of RNA base pairing? A lot less than you may think [new]
RNA pairing learned w/minimal params, seqs only. No aligns/struct.

Influ-BERT: An Interpretable Model for Enhancing Low-Frequency Influenza A virus Subtype Recognition

Figure 1

Figure 3

Figure 4

Influ-BERT: An Interpretable Model for Enhancing Low-Frequency Influenza A virus Subtype Recognition [new]
Domain-adaptive pretraining boosts rare IAV subtype ID; key genomic regions, custom tokenizer, 2-stage training.

Figure 1

Figure 2

Figure 3

Table 2

Tripleknock: predicting lethal effect of three-gene knockout in bacteria by deep learning [new]
Deep learn predict bact. 3-gene lethal knockouts.

Atom-level generative foundation model for molecular interaction with pockets

Figure 1

Figure 2

Figure 3

Atom-level generative foundation model for molecular interaction with pockets [updated]
Gen. model learns atom-level protein-pocket interact. Unified framework handles mol/peptide design & predict w/o fine-tuning.