sequence-to-function modeling

Learning sequence-function relationships with scalable, interpretable Gaussian processes

The paper introduces interpretable Gaussian process models for sequence-function relationships, capturing epistasis and achieving superior predictive performance on large biological datasets.

Antibody immunogenicity prediction and optimization with ImmunoSeq

ImmunoSeq predicts antibody immunogenicity by analyzing k-mer peptides from self and non-self libraries, identifying immunogenic hotspots and optimizing sequence humanness.

Integrating experimental feedback improves generative models for biological sequences

The paper improves generative models for biological sequences by integrating experimental feedback, increasing functional sequence generation from 6.7% to 63.7% in ribozyme design.

Reinforcement Learning for Antibody Sequence Infilling

This paper introduces a reinforcement learning framework for antibody sequence design that improves biophysical properties using infilling models and online learning strategies.

LICHEN: Light-chain Immunoglobulin sequence generation Conditioned on the Heavy chain and Experimental Needs

LICHEN generates light-chain antibody sequences conditioned on heavy chains and experimental needs, producing valid, diverse, and functional sequences.

An open structure-guided RNA foundation model for generalizable prediction

This paper introduces structRFM, a structure-guided RNA foundation model that improves RNA function prediction and structural inference through pre-training on sequence-structure pairs.

DNA Sequence Trace Reconstruction Using Deep Learning

DNARetrace reconstructs DNA sequences from raw data using a deep learning model, achieving competitive results across sequencing methods and downstream tasks.

A novel NGS-compatible Enzymatic Strategy Enables Carryover Contamination Removal and Enhances Sequencing Performance.

This paper introduces a novel enzymatic strategy using 7-deaza-dGTP and Fpg to remove carryover contamination and improve sequencing performance in NGS.

Synteny-aware functional annotation of bacteriophage genomes with Phynteny

Phynteny combines ESM2 protein embeddings, a BiLSTM, and circular-attention transformer layers to exploit phage gene order, assigning one of nine PHROG functional categories to unannotated genes.

Pseudo-perplexity in One Fell Swoop for Protein Fitness Estimation

The paper introduces a method to estimate protein fitness using pseudo-perplexity from a single forward pass through a language model, achieving state-of-the-art results on a benchmark.

DrugPT: A Flexible Framework for Integrating Gene and Chemical Representations in Perturbation Modeling

DrugPT predicts gene expression changes after drug treatment using a Transformer-based framework integrating gene and chemical representations.

Efficient Seeding for Error-Prone Sequences with SubseqHash2

This paper introduces SubseqHash2, an improved subsequence-based seeding method that achieves faster computation while maintaining high accuracy for error-prone long reads.

Missense3D-PTMdb an interactive web resource to explore and visualize genetic variants and post-translational modifications sites (PTMs) using AlphaFold 3D models

Missense3D-PTMdb maps missense variants and PTM sites to AlphaFold 3D models to explore their structural and functional impacts.

From Scarce Functional Labels to Label-Aware Generation in Homologous Protein Families

The paper develops a two-stage strategy for functional annotation and label-aware sequence generation in protein families using limited labels and phylogeny-aware evaluation.

A contextualised protein language model reveals the functional syntax of bacterial evolution

Trained on over 1.3 million bacterial genomes represented as ordered protein sequences, the transformer foundation model Bacformer predicts protein interactions, operons, essentiality, and phenotypes while generating trait-conditioned synthetic proteomes.

Learning the native-like codons with a 5'UTR and RNA secondary structure aided species-informed transformer model

TransCodon is a Transformer-based model that predicts optimal synonymous codons by integrating species-specific information, 5'UTRs, and RNA structure, improving translational efficiency in heterologous hosts.

Scalable emulation of protein equilibrium ensembles with generative deep learning

BioEmu, a diffusion model trained on structures, molecular dynamics trajectories, and stability assays, samples protein equilibrium ensembles and folding free energies within ~1 kcal/mol in GPU-hours.

EZpred: improving deep learning-based enzyme function prediction using unlabeled sequence homologs

EZpred improves enzyme function prediction by using unlabeled sequence homologs, achieving higher F1-scores than existing methods.

Temporal foundation model unveils ancient HepatitisB virus evolution

This paper develops ViraChron, a temporal foundation model for HBV evolution, enabling accurate sequence completion and analysis of ancient HBV genomes.

Medium-sized protein language models perform well at transfer learning on realistic datasets

Medium-sized protein language models (e.g. ESM C 600M) with mean embeddings nearly match ESM-2 15B and ESM C 6B in transfer learning on mutational scanning and PISCES data.

Characterization of selective pressures acting on protein sites with Deep Learning

This paper develops a deep learning model to identify selective pressures on protein sites, showing it outperforms likelihood-based methods on similar data but is less effective on divergent datasets.

SARITA: A Large Language Model for Generating the S1 Subunit of the SARS-CoV-2 Spike Protein

SARITA is a large language model that generates synthetic SARS-CoV-2 Spike S1 sequences, accurately predicting mutations in future variants like Delta and Omicron.

Exploiting uniqueness: seed-chain-extend alignment on elastic founder graphs

This paper introduces a seed-chain-extend alignment workflow using elastic founder graphs to improve sequence-to-graph alignment accuracy and efficiency in pangenomics.

Context-aware geometric deep learning for RNA sequence design

This paper introduces RISoTTo, a geometric deep learning method for RNA sequence design that incorporates molecular context, showing improved accuracy and binding affinity in in silico tests.

A Unified Framework for Model-Informed and Agentic RNA Design

This paper introduces a framework for model-informed mRNA design using integrated gradients and an agentic hypothesis engine, achieving improved in vivo expression.

Frequency-Blended Diffusion Models for Synthetic Generation of Biologically Realistic Splice Site Sequences

This paper introduces a frequency-blended diffusion model to generate biologically realistic splice site sequences, showing improved motif conservation and model transferability.

Benchmarking Inverse Folding Models for Antibody CDR Sequence Design

This study benchmarks AI models for antibody CDR sequence design, finding that specialized models like AntiFold perform better in complex regions while general models struggle with antibody-specific nuances.

Movi Color: fast and accurate long-read classification with the move structure

Movi Color improves long-read classification accuracy and speed using a move structure and color-based indexing, outperforming existing methods in precision and recall.

Nicking Loop™: An infinite amplification of single-stranded circular DNA

The paper introduces Nicking Loop, a method for converting DNA into circular single-stranded DNA for NGS, showing improved reproducibility, sensitivity, and compatibility with sequencing platforms.

DIRseq: a method for predicting drug-interacting residues of intrinsically disordered proteins from sequences

DIRseq predicts drug-interacting residues in intrinsically disordered proteins from sequences, showing good agreement with experimental methods like NMR and MD simulations.