Estimating Change in Foldability Due to Multipoint Deletions in Protein Structures.

Area of Science:

• * Structural Biology and Bioinformatics
• * Protein Engineering and Design

Background:

• * Amino acid insertions/deletions in proteins can alter structure and function, impacting evolution or causing disease.
• * Multipoint deletions (MPDs) are common but lack computational tools to differentiate beneficial from harmful variants.
• * Existing resources do not address the prediction of foldability changes due to MPDs.

Purpose of the Study:

• * To develop a predictive computational framework for assessing the impact of MPDs on protein foldability.
• * To differentiate between beneficial and potentially harmful MPDs in protein structures.
• * To establish a foundation for novel protein engineering applications.

Main Methods:

• * Introduction of a novel dataset comprising 153 beneficial MPD instances and 7650 unlabeled MPD instances.
• * Development of a positive unlabeled learning-based prediction framework (PROFOUND).
• * Utilizing fold-level, environment-specific, and deletion site-specific attributes for prediction.

Main Results:

• * PROFOUND achieved high recall (82.2% for loop, 86.6% for non-loop regions) with low fallout rates (14.2% for loop, 20.6% for non-loop).
• * The low fallout rate indicates non-random effects of MPDs on protein foldability.
• * Inclusion of evolutionary attributes further improved prediction accuracy.

Conclusions:

• * PROFOUND is the first system to predict protein foldability changes due to MPDs.
• * The developed framework and dataset can significantly aid in protein engineering.
• * Understanding MPD impact is crucial for both evolutionary studies and therapeutic development.