DeepFam: deep learning based alignment-free method for protein family modeling and prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Next-generation sequencing generates vast numbers of protein sequences requiring functional annotation.
• Experimental methods for protein function assignment are costly and time-consuming.
• Existing computational methods like k-mer and profile Hidden Markov Models (pHMM) have limitations in accuracy or speed.

Purpose of the Study:

• To develop a more accurate and faster computational method for protein function prediction.
• To address the limitations of current alignment-free and alignment-based approaches.

Main Methods:

• Introduction of DeepFam, a novel alignment-free deep learning method.
• DeepFam extracts functional information directly from protein sequences, bypassing the need for multiple sequence alignments.
• The method was evaluated on Clusters of Orthologous Groups (COGs) and G protein-coupled receptor (GPCR) datasets.

Main Results:

• DeepFam demonstrated superior performance in both accuracy and runtime compared to state-of-the-art methods.
• The method effectively captures conserved regions for protein family modeling.
• DeepFam successfully identified documented conserved regions in the Prosite database during function prediction.

Conclusions:

• DeepFam provides a highly accurate and efficient solution for protein function prediction.
• The deep learning approach is well-suited for characterizing the rapidly growing number of protein sequences.
• The availability of the code facilitates its application in biological research.