Rossmann-toolbox

Rossmann-toolbox predicts and designs cofactor specificity in Rossmann fold proteins by applying deep learning to sequence and structural features of the conserved βαβ motif.

Key Features:

• Cofactor scope: Predicts specificity for nucleoside-based cofactors S-adenosylmethionine (SAM), nicotinamide cofactors (NAD, NADP), and flavin adenine dinucleotide (FAD).
• Motif-focused analysis: Centers predictions and designs on the conserved βαβ motif that determines cofactor binding in Rossmann fold proteins.
• Model architecture: Implements two complementary deep learning models trained on sequence and structural features of the βαβ motif.
• Benchmarking: Validated on independent test sets including motifs dissimilar to the training set and 38 experimentally confirmed rational cofactor design cases.
• Performance: Demonstrates near‑perfect predictive performance on independent benchmarks.
• Biological targets: Addresses cofactor interactions relevant to Rossmann methyltransferases and oxidoreductases implicated in nucleotide and amino acid metabolism.

Scientific Applications:

• Cofactor specificity prediction: Enables assignment of cofactor preference for Rossmann fold βαβ motifs among SAM, NAD, NADP, and FAD.
• Rational cofactor design: Supports redesign of βαβ motifs to alter cofactor specificity, validated by 38 experimental cases.
• Enzyme engineering for metabolism: Facilitates computational engineering of methyltransferases and oxidoreductases to modify cofactor usage in nucleotide and amino acid metabolic contexts.

Methodology:

Two complementary deep learning models were trained on sequence and structural features of the βαβ motif and benchmarked against independent test sets—including motifs with no resemblance to the training set and 38 experimentally confirmed cofactor redesign cases—reporting near‑perfect performance.