large language model

LPT: Orchestrating bounded agentic loops for automated protein binder design

LPT uses bounded agentic loops with LLMs and a curated database to design protein binders from natural language prompts, achieving high success rates with low computational costs.

Toward De Novo Protein Design from Natural Language

This paper introduces Pinal, a foundation model that designs proteins from natural language descriptions, achieving functional alignment and experimental success in four design tasks.

Reliable and Efficient Agentic Protein Design with Toolset Harnessing

Chaperone, a DataFrame-based framework and agent harness for protein design tools, cut tokens 86% and GPU-hours 64% while raising pass rate from 6% to 12%.

In Other Words: Learning When to Trust Generated Protein Function Descriptions

The paper introduces a method to assess the reliability of generated protein function descriptions using a judge model and shows it improves performance across multiple methods.

What comes after de novo? Automated lead optimization of proteins with CRADLE-1

CRADLE-1 automates protein lead optimization, improving speed and efficiency compared to rational design methods.

Inherent limitations of zero-shot mutation prediction in protein engineering

This study evaluates zero-shot mutation prediction models in protein engineering, finding they struggle to identify optimal mutations and predict new functions due to limitations in capturing evolutionary constraints.

EdiProPred: Prediction of Edible Plant Tissue-Associated Proteins Using Large Language Models

The paper develops EdiProPred, a model using large language models to classify plant proteins as edible or non-edible based on sequence data, achieving high accuracy.

Evidence Scaling for Zero-Shot Protein Reasoning with Large Language Models

This paper shows that adding biological evidence to large language models improves zero-shot protein variant prediction without retraining the model.

AnewDDE: An Agentic Drug Discovery Engine for Biomolecular Interaction Modelling and Closed-Loop Design

AnewDDE integrates AI tools for biomolecular interaction modelling and drug design, demonstrating improved performance in structure prediction, binder identification, affinity estimation, and ADMET prediction.

Generative Language Modeling for Antibody CDR Grafting and Alignment-driven De Novo Design

GenCDR uses LLaMa-based models to generate antibody CDRs with framework context, achieving high recovery and developable designs through alignment-driven training.

De novo Rubisco design with protein language models

This paper designs novel Rubisco variants using protein language models, identifying active candidates with improved CO2/O2 specificity and structural validity.

De novo design of ligand binding proteins using large language models alone

This paper demonstrates that large language models can generate de novo protein sequences that bind specific ligands without copying existing sequences, with a 25% hit rate in experimental validation.

RegimeFormer: A Large Protein Model of Global Perturbation Regimes

RegimeFormer identifies protein perturbation regimes using a large-scale model and atlas, improving prediction accuracy and enabling downstream applications.

BioReason-Pro: Advancing Protein Function Prediction with Multimodal Biological Reasoning

A reasoning LLM combining ESM3 embeddings with GO-GPT term predictions produces protein function annotations that experts rated tie-or-better than UniProt in 79% of cases (SFT model).

VirProtRAG: Literature-grounded viral protein function annotation with retrieval-augmented generation

VirProtRAG improves viral protein function annotation by integrating literature retrieval and evidence-based generation, introducing 32.53% new annotations and supporting sequence-inferred functions.

Agentomics: An Agentic System that Autonomously Develops Novel State-of-the-art Solutions for Biomedical Machine Learning Tasks

Agentomics is an autonomous system that generates ML models for biomedical tasks, outperforming existing agentic systems and human-engineered models on 11/20 datasets.

OmniGene-4: A Unified Bio-Language MoE Model with Router-Level Interpretability

OmniGene-4 is a multi-modal bio-language MoE model that decomposes task performance by router states, achieving strong benchmarks in homology and biological-knowledge tasks while being computationally efficient.

Bio-BLIP: A Multimodal Architecture for Transferable Reasoning in Genomic Variant Interpretation

Bio-BLIP is a multimodal architecture that integrates DNA, genes, proteins, and text to improve variant interpretation without task-specific fine-tuning, achieving better performance in variant prioritization and target gene prediction.

Aiki-GeNano: Multi-Stage Preference Optimization for Generative Design of Developable Nanobodies

Aiki-GeNano improves nanobody design by integrating developability signals into training, enhancing thermal stability and reducing liabilities while preserving humanness.

yakRNA Design: A semantic multimodal RNA composer

The paper introduces yakRNA Design, a language model that generates functional synthetic RNAs without 3D structural data, achieving efficient frame-shifting capabilities in ribosomes.

Co-designing sequence and structure of functional de novo enzymes with EnzyGen2

EnzyGen2 co-designs protein sequences and structures with ligand-guided functionality, outperforming existing methods in in silico benchmarks and generating active enzymes with high novelty.

Large Language Model Agent for Modular Task Execution in Drug Discovery

AgentD, a GPT-4o agent wired to retrieval, generative, ADMET and Boltz-2 tools, chains early drug-discovery steps for BCL-2 and raised molecules with QED > 0.6 from 34 to 55.

Moremi Bio Agent: Using Neisseria meningitidis Reference Data For The Double Blinded Validation of A General Purpose Biology-Trained Reasoning Model for Pathogen and Antigen Discovery

The paper validates a general-purpose biology-trained AI model for predicting antibody antigen targets using only amino acid sequences, achieving correct predictions in three out of ten tests.

A Unified Multi-modal LLM for Dynamic Protein-Ligand Interactions and Generative Molecular Design

BioDynaGen unifies protein sequences, ligand SMILES, and text into a single model for predicting binding and generating molecules.

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.

Genolator: A Multimodal Large Language Model Fusing Natural Language, Genomic, and Structural Tokens for Protein Function Interpretation

Genolator is a multimodal large language model that combines genomic and structural data with natural language to interpret protein function, outperforming existing models in accuracy.

PLiCat: Decoding protein-lipid interactions by large language model

PLiCat uses large language models to classify protein-lipid interactions across eight lipid categories and predict lipid-binding sites and mutation impacts.

Designing proteins with reduced T-cell epitopes through policy optimization

The paper designs proteins with reduced T-cell epitopes using deep generative models and policy optimization to improve immune compatibility across MHC Class I alleles.

Conserved heavy/light contacts and germline preferences revealed by a large-scale analysis of natively paired human antibody sequences and structural data.

PairedAbNGS pools 58 single-cell bioprojects of natural paired human and mouse antibodies, revealing heavy/light V gene pairing biases beyond usage frequencies and conserved CDR3-FWR2 inter-chain contacts.

Reading papers: Extraction of molecular interaction networks with large language models

The paper uses large language models to extract molecular interaction networks from scientific texts, achieving high precision and recall in protein-protein and gene regulatory network relations.