sequence-to-function modeling

ProteinMPNN Recovers Complex Sequence Properties of Transmembrane β-Barrels

This study benchmarks ProteinMPNN for designing transmembrane β-barrel proteins, showing it outperforms Franklin2023 with refined inputs and generates sequences suitable for experimental validation.

Target-aware Molecule Generation for Drug Design Using a Chemical Language Model

This paper introduces TamGen, a chemical language model for target-aware molecule generation, and shows it produces high-quality compounds with inhibitory activity against Tuberculosis ClpP protease.

Influenza sequence validation and annotation using VADR

The paper introduces VADR, a standalone tool for validating and annotating influenza sequences, showing consistent results with FLAN and extending its capabilities to new subtypes.

WAS IT A MATch I SAW? Approximate palindromes lead to overstated false match rates in benchmarks using reversed sequences

The paper shows that reversed biological sequences in benchmarks can overstate false match rates due to palindromic structures, affecting sensitivity and tool evaluation.

Peptide generative design with weakly order-dependent autoregressive language model and lifelong learning

This paper introduces PepGenWOA, a generative model for bioactive peptide design that outperforms existing methods and enables efficient virtual screening.

Unexplored regions of the protein sequence-structure map revealed at scale by a library of foldtuned language models

This paper introduces foldtuning, a method to generate novel protein sequences with target structures, bypassing evolutionary barriers and enabling synthetic biology applications.

Structure-conditioned masked language models for protein sequence design generalize beyond the native sequence space

This paper introduces Frame2seq, a structure-conditioned masked language model for protein sequence design that generates novel, stable proteins with low sequence identity to native sequences.

MIMt: A curated 16S rRNA reference database with less redundancy and higher accuracy at species-level identification

The paper introduces MIMt, a curated 16S rRNA database that improves species-level identification accuracy and reduces redundancy compared to existing databases.

Targeted Amplification and Genetic Sequencing of the Severe Acute Respiratory Syndrome Coronavirus 2 Surface Glycoprotein

The paper develops SpikeSeq, a targeted sequencing method for the SARS-CoV-2 S-gene, showing high conservation and accuracy across variants.

A cross-attention transformer encoder for paired sequence data

The paper introduces a cross-attention transformer encoder for multi-sequence inputs and evaluates it on peptide-MHCII binding prediction.

ProteinGym: Large-Scale Benchmarks for Protein Design and Fitness Prediction

ProteinGym unifies over 250 deep mutational scans and ~65k clinical variants to benchmark over 70 models in zero-shot and supervised settings; TranceptEVE ranks best overall.

Hybrid-hybrid correction of errors in long reads with HERO

HERO improves error correction in long reads by combining de Bruijn and overlap graphs, reducing indel and mismatch errors by 65% and 20% respectively.

Differentiable Partition Function Calculation for RNA

This paper introduces a differentiable version of McCaskill's RNA partition function algorithm, enabling gradient-based RNA sequence design for applications like mRNA vaccines.

Language model-based B cell receptor sequence embeddings can effectively encode receptor specificity

This study evaluates B cell receptor sequence embeddings for predicting sequence properties and specificity, finding that BCR-specific embeddings and paired-chain sequences improve performance.

Cell-free biosynthesis combined with deep learning accelerates de novo-development of antimicrobial peptides

Autoencoder design plus a roughly 24-hour cell-free screen tested 500 candidates over five rounds, finding 30 functional AMPs; six were broad-spectrum with no resistance emerging in E. coli.

Rational design of disordered proteins for systematic sequence-to-function investigation

This paper introduces GOOSE, a computational framework for designing intrinsically disordered proteins, and demonstrates its ability to create IDRs with specific functional responses.

Protein Language Model Supervised Scalable Approach for Diverse and Designable Protein Motif-Scaffolding with GPDL

The paper introduces GPDL, a protein language model approach for motif-scaffolding that outperforms existing methods in generating diverse and accurate protein structures.

Pervasive, conserved secondary structure in highly charged protein regions

The paper shows that highly charged protein regions often form structured helices, challenging the assumption that they are disordered, and introduces a better classifier for predicting structure in these regions.

GUANinE v1.0: Benchmark Datasets for Genomic AISequence-to-Function Models

GUANinE v1.0 provides benchmark datasets for evaluating genomic AI sequence-to-function models across functional genomics and evolutionary biology tasks.

AP-LASR: Automated Protein Libraries from Ancestral Sequence Reconstruction

AP-LASR automates ancestral sequence reconstruction and generates diverse protein libraries by leveraging ASR uncertainty, improving eLnP scores over existing methods.

New distance measure for comparing protein using cellular automata image

The paper introduces a new distance measure using cellular automata to compare protein sequences and identify common ancestors across species.

The amino acid sequence determines protein abundance through its conformational stability and reduced synthesis cost.

This study shows that amino acid sequences determine protein abundance by affecting conformational stability and synthesis cost, using deep learning and yeast experiments.

Protein generation with evolutionary diffusion: sequence is all you need

This paper introduces EvoDiff, a diffusion model that generates diverse, high-fidelity proteins in sequence space, overcoming limitations of structure-based models.

Learning the differences: a transfer-learning approach to predict antigen immunogenicity and T-cell receptor specificity

This paper introduces diffRBM, a transfer-learning model using Restricted Boltzmann Machines to predict antigen immunogenicity and T-cell receptor specificity, achieving performance comparable to existing methods.

SUMOnet: Deep Sequential Prediction of SUMOylation Sites

The paper introduces three deep learning models for predicting SUMOylation sites and reports that SUMOnet-3 outperforms existing methods with 75.8% AUPR and 87% AUC.

Minmers are a generalization of minimizers that enable unbiased local Jaccard estimation

The paper introduces minmers, a generalization of minimizers that provide unbiased local Jaccard similarity estimates, improving the accuracy and speed of sequence comparison tools like MashMap.

PandoraGAN: Generating antiviral peptides using Generative Adversarial Network

PandoraGAN generates novel antiviral peptides using a GAN model trained on curated data, producing sequences with properties similar to known active peptides.

A conditional protein diffusion model generates artificial programmable endonuclease sequences with enhanced activity

This paper introduces CPDiffusion, a conditional protein diffusion model that generates artificial endonuclease sequences with enhanced activity without labeled data.

Structure-based self-supervised learning enables ultrafast prediction of stability changes upon mutation at the protein universe scale

This paper introduces Pythia, a self-supervised graph neural network for fast and accurate prediction of mutation-induced stability changes across protein structures, showing superior performance and speed.

Efficient and accurate sequence generation with small-scale protein language models

This paper introduces a small-scale protein language model that achieves comparable performance to larger models with significantly reduced computational requirements.