MENDELSEEK: An algorithm that predicts mendelian genes and elucidates what makes them special

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Mendelian diseases, caused by single gene defects, have a significant collective burden.
• Identifying causative genes is crucial for diagnosis and treatment, yet over half remain unknown.
• Existing methods for gene identification have limitations.

Purpose of the Study:

• To develop and validate MENDELSEEK, a novel machine learning framework for predicting Mendelian genes.
• To improve the accuracy and efficiency of identifying genes responsible for Mendelian diseases.
• To uncover distinguishing features of Mendelian genes.

Main Methods:

• MENDELSEEK integrates residue variation scores, pathway participation, Gene Ontology (GO) processes, and protein language model features.
• The framework was benchmarked across 16,946 human genes using 10-fold cross-validation.
• Performance was evaluated using Area Under the Curve (AUC) and Area Under the Precision-Recall Curve (AUPR).

Main Results:

• MENDELSEEK achieved a high AUC of 0.869 and AUPR of 0.737, outperforming ENTPRISE+ENTPRISE-X (AUC 0.781; AUPR 0.626) and REVEL (AUC 0.585; AUPR 0.401).
• Applied to all human genes, MENDELSEEK predicted 1,277 novel Mendelian gene candidates with precision > 0.7.
• Mendelian genes exhibit significantly more protein-protein interactions and are evolutionarily ancient compared to non-Mendelian genes.

Conclusions:

• MENDELSEEK represents a significant advancement in predicting Mendelian genes.
• The framework offers a powerful tool for discovering novel disease-gene associations.
• The study provides new insights into the molecular and evolutionary characteristics differentiating Mendelian genes.