The Engine

The AlphaFold engine provides state-of-the-art protein structure prediction using Google DeepMind’s AlphaFold2 architecture, implemented through the optimized ColabFold framework. It predicts accurate 3D protein conformations directly from amino acid sequences, supporting both monomeric and complex structures. This engine enables high-confidence modeling of protein targets for drug discovery, interaction analysis, and structure-based design.

Users enter their amino acid sequence, configure parameters such as relaxation count, template mode, and MSA options, then run a campaign from Revilico’s Command Center. The pipeline outputs fully folded PDB files, residue-level confidence scores, and visualized structures ready for downstream analysis.

The Algorithm

The AlphaFold engine unifies advanced deep learning and evolutionary modeling to infer protein 3D structures from sequence data.

• Sequence Input: Users provide a protein or complex sequence, optionally separated by chains.

• Multiple Sequence Alignment (MSA): Generated using mmseqs2 against UniRef and environmental databases, or custom MSAs if supplied.

• Model Selection: Automatically chooses between AlphaFold2-PTM for monomers and Multimer v3 for complexes.

• Structure Prediction: Predicts folded conformations with per-residue confidence estimates (pLDDT).

• Relaxation: Optionally refines top structures using AMBER energy minimization for improved stereochemical accuracy.

All outputs, including predicted structures, confidence maps, and MSA details, are available through the integrated AlphaFold Analytics dashboard for interactive visualization and export.

Algorithm Validation

The AlphaFold system has demonstrated near-experimental accuracy in global benchmarks such as CASP14, achieving median backbone RMSDs within 1 Å of resolved structures. Internal validations at Revilico confirm that AlphaFold predictions align closely with crystallographic and cryo-EM references for diverse target classes. Confidence metrics (pLDDT and predicted TM-score) reliably correlate with structural accuracy, providing transparent quality assessments for each prediction. Relaxed models consistently improve local geometry without compromising global folds, supporting their use in downstream modeling and docking studies.

Scientific Impact

The AlphaFold engine enables structural analysis at scale, making it possible to:

• Predict high-resolution target structures when no experimental data exists.

• Explore mutation effects and conformational variability in silico.

• Generate structural templates for molecular docking and dynamics simulations.

• Model protein-protein complexes to understand binding interfaces.

• Perform comparative analysis across homologous targets to assess druggability.

This engine transforms sequence information into actionable 3D insight, accelerating structure-based drug discovery and enabling new analyses in structural bioinformatics and protein engineering.

Business Impact

By automating high-accuracy structure prediction, the AlphaFold engine allows teams to:

• Bypass experimental bottlenecks in early target characterization.

• Rapidly generate structures for docking, virtual screening, and lead optimization.

• Integrate structure prediction directly into multi-target campaigns.

• Reduce dependency on costly and time-intensive crystallography or cryo-EM studies.

The result is faster, lower-cost target validation and a more agile discovery pipeline, empowering organizations to move from sequence to structure to drug candidate with unprecedented efficiency.