High-Accuracy Protein Structure Prediction with ESMFold2 is Now Available on Vecura

What is ESMFold2?

ESMFold2 is the successor to ESMFold that sets a new state of the art for single-sequence structure prediction and enables the generation of new functional proteins through searching the ESMC model's latent space. Built on the ESMC 6B model, it combines large language model embeddings with a diffusion-based structure prediction architecture to predict high-resolution, all-atom 3D protein structures directly from amino acid sequences, with optional multiple sequence alignment (MSA) input for enhanced accuracy on challenging targets. It leads across standard protein folding benchmarks for predicting protein-protein and antibody-antigen interactions, outperforming AlphaFold 3 on antibody-antigen binding pose prediction from ESMC representations alone.

It helps users predict the 3D structure of proteins and biomolecular complexes — including protein-protein and antibody-antigen interactions — directly from sequence input. It is especially useful for early-stage therapeutic discovery, particularly antibody engineering and the de novo design of protein binders against disease-relevant targets.

What can users do with ESMFold2 on Vecura?

With ESMFold2 on Vecura, users can:

• Submit a protein sequence and receive a predicted all-atom 3D structure used as structural input for downstream DTIGN-based drug-target interaction modeling

• Run fast, single-sequence folding without MSA input — ideal for novel, metagenomic, or poorly characterized targets

• Review confidence scores (pLDDT, PAE) to assess which structural regions are reliable enough for virtual screening and docking

• Feed predicted structures directly into Vecura's virtual screening workflow, enabling a seamless path from sequence → structure → compound ranking

What the output means

The output provides comprehensive structural information including all-atom coordinates (backbone and side chains), confidence metrics, and optional distogram predictions for detailed analysis of predicted structures. Within Vecura, this structural output serves as the target input for DTIGN-based drug-target interaction prediction and virtual screening. High-confidence regions (high pLDDT score) are suitable for binding site identification and docking; low-confidence regions should be deprioritized or flagged for experimental follow-up.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

In traditional drug discovery, researchers must screen millions of molecules to find those that act on the right biological target — a process that is slow because it requires ruling out the many that are unsafe, unstable, or ineffective. Accurate 3D protein structure is the foundation of structure-based virtual screening: without it, predicting how a drug molecule fits a target is guesswork. By embedding ESMFold2 into Vecura's workflow, users can now computationally resolve a target's structure and immediately proceed to compound screening — all within one platform.

VECURA integrates cutting-edge metagenomics and metaproteomics analysis with a vast Natural Compound Library from tropical flora, enabling rapid identification of promising drug candidates and significantly reducing discovery time and costs. ESMFold2 extends this capability to protein targets that have never been experimentally crystallized, dramatically widening the scope of what Vecura's pipeline can tackle.

• Developed by: Biohub (led by Alex Rives)

• Source: Official GitHub Repository

• Reference: Language Modeling Materializes a World Model of Protein Biology