MEMO-Stab2: Multi-View Sequence-Based Deep Learning Framework for Predicting Mutation-Induced Stability Changes in...

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Accurate prediction of protein thermodynamic stability changes due to point mutations is vital for understanding protein function and engineering proteins.
• Transmembrane proteins (TMPs) are critical for cellular processes and drug development but are challenging to study due to limited structural data.
• Current prediction tools often require 3D structures or multiple sequence alignments, which are frequently unavailable or of poor quality for TMPs.

Purpose of the Study:

• To develop a fast, structure-independent deep learning framework for predicting mutation-induced stability changes in TMPs.
• To overcome the limitations of existing predictors by not requiring experimental 3D structures or explicit multiple sequence alignments.

Main Methods:

• Introduced MEMO-Stab2, a deep learning framework that reformulates mutation stability prediction as a binary classification task.
• Integrated multiview features using a Transformer architecture, incorporating embeddings from multiple pretrained protein language models (PLMs) and PLM-based structural predictions.
• Leveraged PLMs to implicitly capture evolutionary and structural information directly from amino acid sequences.

Main Results:

• MEMO-Stab2 achieved high performance, with an F1 score of 0.92 on an internal benchmark, outperforming existing specialized and general prediction tools.
• Demonstrated robust generalization across diverse TMP families with low sequence identity and superior performance in challenging regions like the transmembrane core.
• Validated computational efficiency, enabling large-scale mutation screening within minutes.

Conclusions:

• MEMO-Stab2 provides a practical, robust, and efficient solution for predicting mutation effects on TMP stability.
• The structure-independent approach significantly expands the applicability of stability prediction to TMPs.
• This tool facilitates enhanced transmembrane protein variant evaluation and engineering efforts.