Two-dimensional geometric template diffusion for boosting single-sequence protein structure prediction

2026-04-01 · Source: Nature Machine Intelligence · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences · Depth: Expert, long

Summary

TDFold, a novel two-dimensional geometric template diffusion method, significantly enhances single-sequence protein structure prediction by generating high-quality pairwise geometries. This method, detailed in Nature Machine Intelligence (2026), employs a two-stage network architecture: 2D geometric template generation followed by sequence-geometry collaborative learning. TDFold outperforms existing protein language models like ESMFold and OmegaFold, as well as homology-based methods such as AlphaFold2, AlphaFold3, and RoseTTAFold, in terms of prediction accuracy, resource efficiency, and inference speed. Its effectiveness is demonstrated on homology-insufficient datasets like Orphan and Orphan25, alongside standard CASP benchmarks (CASP14, CASP15, CASP16), offering a viable alternative for resource-constrained research institutions.

Key takeaway

For AI Scientists and Research Scientists focused on protein structure prediction, TDFold offers a compelling alternative to homology-based and existing language models. Its superior performance on single-sequence data, coupled with lower resource consumption and higher inference efficiency, makes it particularly valuable for accelerating research in resource-limited environments. Consider integrating TDFold into your computational pipeline for more accurate and efficient structural insights, especially for novel or orphan proteins.

Key insights

TDFold uses 2D geometric template diffusion for efficient, accurate single-sequence protein structure prediction.

Principles

• Diffusion models can generate high-quality geometric templates.
• Single-sequence prediction reduces computational homology costs.
• Collaborative learning improves structure inference.

Method

TDFold infers 3D protein structure via a two-stage network: 2D geometric template generation and sequence-geometry collaborative learning, using diffusion to create pairwise distances and orientations.

In practice

• Utilize TDFold for protein structure prediction.
• Apply 2D diffusion for geometric template generation.
• Benchmark against Orphan and CASP datasets.

Topics

• TDFold
• Protein Structure Prediction
• Geometric Template Diffusion
• Single-Sequence Prediction
• Computational Efficiency

Code references

• soedinglab/hh-suite
• huggingface/diffusers
• pyg-team/pytorch_geometric
• matplotlib/matplotlib
• numpy/numpy

Best for: AI Scientist, Research Scientist

Related on AIssential

• Move over, AlphaFold: open-source model predicts shape of 1 billion proteins — ESMFold2, an open-source AI, predicts over a billion protein structures, outperforming AlphaFold3 and enabling novel protein design.
• Predicting RNA 3D structure and conformers using a pre-trained secondary structure model and structure-aware attention — trRosettaRNA2 accurately predicts RNA 3D structures and conformers by integrating pre-trained secondary structure priors and SS-aware attention.
• Author Correction: Mask-prior-guided denoising diffusion improves inverse protein folding — A published correction clarifies MapDiff's amino acid prediction accuracy in inverse protein folding.
• Learning Topological Representations for Molecular Dynamics — Masked Flood complex enhances persistent homology for molecular dynamics by incorporating protein-specific structural bias.
• Viral Proteins Reveal Geometry of Protein Language Models — Protein language models organize sequence embeddings by a "nativeness axis" while preserving specific biological group signals.
• How to Accelerate Protein Structure Prediction at Proteome-Scale — Accelerated computing and optimized workflows enable proteome-scale prediction of protein complex structures.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Nature Machine Intelligence.