sequence-to-function modeling

Controllable Protein Design via Autoregressive Direct Coupling Analysis Conditioned on Principal Components

FeatureDCA improves protein sequence generation by conditioning DCA on principal components, achieving accurate structural and functional results across multiple families.

BOND-PEP: topology-conditioned bipartite alignment for evidence-grounded peptide binder generation

BOND-PEP generates controllable peptide binders using topology-conditioned bipartite alignment and retrieval-augmented methods, achieving state-of-the-art results with sequence-only inputs.

Advancing Codon Language Modeling with Synonymous Codon Constrained Masking

This paper introduces SynCodonLM, a codon language model that enforces synonymous codon constraints, improving performance on DNA-level biology benchmarks.

CodonRL: Multi-Objective Codon Sequence Optimization Using Demonstration-Guided Reinforcement Learning

CodonRL improves codon sequence optimization using reinforcement learning with structural feedback and expert demonstration, outperforming existing methods on multiple metrics.

Out-of-equilibrium selection pressure enhances inference from protein sequence data

The paper shows that fluctuating selection pressures improve coevolution-based inference of protein structure and function from sequence data.

Automatically Defining Protein Words for Diverse Functional Predictions Based on Attention Analysis of a Protein Language Model

The paper introduces a toolkit using protein language models to define 'protein words' for functional prediction, outperforming motif-based methods across multiple tasks.

Quantitative profiling of millions of nucleotides reveals sequence-encoded interactions that govern plasmid propagation

The paper develops a sequencing-based framework and a neural network model to predict plasmid propagation from sequence, revealing sequence-encoded interactions that influence plasmid stability and host compatibility.

UdonPred: Untangling Protein Intrinsic Disorder Prediction

UdonPred predicts per-residue protein disorder from sequence using embeddings from ProstT5, showing that annotation quality, not model capacity, drives performance in continuous disorder prediction.

AistSeq: An In-House Tn5-Based Plasmid Sequencing Platform Using A Compact Benchtop Sequencer

This paper describes an in-house Tn5-based plasmid sequencing platform called AistSeq using a compact benchtop sequencer, demonstrating a streamlined workflow for plasmid sequence verification.

HyDrop v2: Scalable atlas construction for training sequence-to-function models

The paper introduces HyDrop v2, a microfluidic system for generating large scATAC-seq atlases, and shows its data is comparable to commercial platforms for training sequence-to-function models.

ColiFormer: A Transformer-Based Codon Optimization Model Balancing Multiple Objectives for Enhanced E. coli Gene Expression

ColiFormer improves E. coli gene expression by balancing multiple codon optimization objectives using a transformer model, outperforming existing methods in CAI, tAI, and cis-regulatory element reduction.

STRUCT2SEQ: RNA INVERSE FOLDING WITH DEEP Q-LEARNING

Struct2SeQ uses deep Q-learning to design RNA sequences that fold into target structures and match SHAPE reactivity constraints, outperforming humans and existing methods.

Multi-objective optimization for designing structurally similar proteins with diverse sequences

Inferring context-specific site variation with evotuned protein language models

This study introduces a pLM entropy metric to assess protein site conservation and variability, showing that fine-tuning pLMs on evolutionarily related proteins improves their ability to capture context-specific site variation.

CyclicMPNN: Stable Cyclic Peptide Sequence Generation

The paper introduces CyclicMPNN, a model that generates energetically stable cyclic peptide sequences by fine-tuning ProteinMPNN with structural and in silico data, improving success rates in folding into generated backbones.

A multispecies foundation model for genome function prediction, gene annotation and sequence design

NTv3 is a foundation model that predicts genome function, annotates genes, and designs DNA sequences across species, outperforming existing models in multiple tasks.

NW-flex: flexible-block sequence alignment for short tandem repeats

NW-flex improves STR read alignment by allowing flexible block contraction while maintaining alignment optimality and using minimal computational overhead.

Large generative mRNA language foundation model for efficient coding sequence generation and design with mRNA-GPT

The paper introduces mRNA-GPT, a generative mRNA language model trained on sequences from bacteria, eukaryotes, and archaea, showing improved translation efficiency and stability in mRNA design.

A Modular Fusion Neural Network Approach to Efficiently Predict Multi-Metal Binding Sites in Protein Sequences

This paper introduces a modular deep learning framework to predict multi-metal binding sites in protein sequences, improving accuracy and efficiency for large-scale proteome annotation.

Pangenome-Informed Language Models for Synthetic Genome Sequence Generation

This paper introduces pangenome-informed language models for generating synthetic DNA sequences, showing improved efficiency and quality over classical methods.

Protein Structure Informed Bacteriophage Genome Annotation with Phold

Phold pairs ProstT5-predicted 3Di tokens with Foldseek searches against 1,363,704 curated phage structures, annotating 49.4% of CDS in 1,419 INPHARED genomes versus 34.7% for Pharokka MMseqs2.

Rewriting protein alphabets with language models

This paper introduces a new 20-letter alphabet, TEA, derived from protein language models, enabling efficient homology searches comparable to structure-based methods without structural data.

LCSKPOA: Enabling banded semi-global partial order alignments via efficient and accurate backbone generation through extended lcsk++

The paper introduces an extended LCSk++ algorithm for graph-based alignment, improving memory and runtime efficiency for local and semi-global POA while maintaining alignment quality.

ClsDiff-AMP30: Generating Antimicrobial Peptides by a Classifier Guidance Noise Predictor

ClsDiff-AMP30 generates short antimicrobial peptides with high antimicrobial activity and low toxicity through a classifier-guided diffusion model.

Helixer--de novo Prediction of Primary Eukaryotic Gene Models Combining Deep Learning and a Hidden Markov Model

Helixer, a hybrid convolutional/bLSTM network with a hidden Markov model post-processor, predicts eukaryotic primary gene models from DNA alone, outscoring GenemarkES and AUGUSTUS on most species.

CodonTranslator: a conditional codon language model for codon optimization across all domains of life

CodonTranslator is a Transformer-based model for codon optimization across species, achieving better codon usage and stability than existing methods.

General-Purpose Large Language Models, such as DeepSeek V3.2, Have Evolved Protein Design Capabilities

This study shows that general-purpose LLMs like DeepSeek V3.2 can generate protein sequences with desired structures and improve solubility without biological training.

BiRNA-BERT Allows Efficient RNA Language Modeling with Adaptive Tokenization

BiRNA-BERT is a Transformer-based model pretrained on RNA sequences using adaptive tokenization, achieving state-of-the-art results in long-sequence tasks and efficient performance on short sequences.

STAMPS: Signal-peptide Transformer for Augmenting Mammalian Protein Secretion

STAMPS designs novel mammalian signal peptides to enhance protein secretion, outperforming existing benchmarks in multiple cell lines.

SeqForge: A scalable platform for alignment-based searches, motif detection, and sequence curation across meta/genomic datasets

SeqForge provides a scalable toolkit for alignment-based searches, motif detection, and sequence curation in metagenomic datasets, reducing the need for custom scripting.