FeSseqdb: a curated sequence-level database and interpretable machine learning framework for identifying iron-sulfur...

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Iron-sulfur (Fe-S) clusters are vital cofactors in metalloproteins, essential for biological processes like electron transfer and catalysis.
• Identifying Fe-S proteins is challenging due to experimental limitations and database inconsistencies.
• Accurate identification is crucial for understanding Fe-S protein functions and their roles in various biological pathways.

Purpose of the Study:

• To develop a reliable resource for Fe-S protein research.
• To create a machine learning framework for predicting Fe-S proteins using only sequence data.
• To uncover sequence features critical for Fe-S protein identification and function.

Main Methods:

• Creation of FeSseqdb, a curated database of Fe-S cluster-containing proteins from PDB, verified by atomic coordinates.
• Development of a machine learning model using sequence-derived features (amino acid composition, cysteine metrics, length).
• Application of explainable AI to identify key sequence characteristics driving Fe-S protein prediction.

Main Results:

• FeSseqdb provides a standardized and reliable resource for Fe-S protein research.
• A random forest model achieved high predictive performance for Fe-S proteins using sequence features.
• Explainable AI identified cysteine frequency/distribution and proline content as primary predictors, with serine, glutamic acid, and arginine as secondary determinants.

Conclusions:

• Sequence-based machine learning models can accurately predict Fe-S proteins.
• These models offer insights into the biochemical principles governing Fe-S protein structure and function.
• This approach facilitates large-scale, sequence-based discovery of Fe-S proteins and their functional diversity.