SAAFEC-SEQ: an online application for calculating folding free energy changes in proteins caused by missense mutations

Answer: SAAFEC-SEQ predicts the effect of single amino-acid substitutions on protein thermodynamic stability. Specifically, it predicts the mutation-induced change in protein folding free energy, ΔΔG, using only the protein sequence.

Answer: SAAFEC-SEQ is a sequence-based machine-learning method. It uses a gradient boosting decision tree algorithm to predict changes in protein folding free energy caused by single-point missense mutations. Unlike structure-based methods, SAAFEC-SEQ does not require a three-dimensional protein structure.

Answer: No. SAAFEC-SEQ requires only the amino-acid sequence of the protein. This makes it useful for proteins without experimentally solved structures and for large-scale variant analysis.

Answer: In SAAFEC-SEQ, ΔΔG is defined as: where ΔGWT is the folding free energy of the wild-type protein and ΔGMT is the folding free energy of the mutant protein.

Folding free energy is often a negative quantity. A more negative folding free energy generally indicates a more favorable and more stable folded state, while a less negative value indicates reduced stability.

Answer: Because SAAFEC-SEQ uses the convention ΔΔG = ΔGWT - ΔGMT. For example: In this example, the mutant has a less negative folding free energy than the wild type. Therefore, the mutant folded state is less favorable, and the mutation is predicted to be destabilizing.

With the SAAFEC-SEQ convention:

• Negative ΔΔG indicates a destabilizing mutation.
• Positive ΔΔG indicates a stabilizing mutation.

Answer: The predicted ΔΔG is reported in kcal/mol.

Answer: SAAFEC-SEQ is designed for single-point missense mutations, where one amino acid in the protein sequence is substituted by another amino acid.

Answer: For a single-mutation prediction, users should provide:

• A protein sequence in FASTA format, either uploaded as a FASTA file or pasted into the input box.
• The wild-type amino acid.
• The mutation position in the submitted sequence.
• The mutant amino acid.

Answer: Yes. In batch mode, users can submit one protein sequence and a mutation-list file containing one single-point mutation per line. Each mutation should be specified relative to the submitted protein sequence.

Answer: SAAFEC-SEQ uses sequence-derived and knowledge-based features, including:

• Physicochemical properties of the mutation site.
• Sequence-neighbor features around the mutation position.
• Evolutionary information from position-specific scoring matrix based features.
• Neighbor mutation conservation scores around the mutation site.

Answer: PsePSSM stands for pseudo-position specific scoring matrix. It is used to capture evolutionary information and sequence-order information from the protein sequence. In SAAFEC-SEQ, PsePSSM-based features are among the most important feature groups for prediction.

Answer: Neighbor mutation conservation scores describe the evolutionary conservation of the mutation site and nearby residues. SAAFEC-SEQ considers residues surrounding the mutation site to capture how conserved the local sequence environment is.

Answer: The method uses physicochemical properties related to the mutation site, such as volume, hydrophobicity, flexibility, chemical property, size, polarity, hydrogen-bonding behavior, and mutation type.

Answer: SAAFEC-SEQ was trained using the S2648 dataset, which contains 2648 experimentally measured single-point missense mutations from 131 proteins. These data were collected from the ProTherm database.

Answer: SAAFEC-SEQ was evaluated using repeated five-fold cross-validation on the S2648 dataset and was further tested on independent datasets, including S350, S276, p53, PTEN, TPMT, and the CAGI5-related datasets.

Answer: In the SAAFEC-SEQ study, the method achieved a Pearson correlation coefficient of approximately 0.75 and a mean squared error of approximately 0.95 kcal/mol on the S2648 dataset using repeated five-fold cross-validation. It also performed competitively or better than several other sequence-based methods on independent benchmark datasets.

Answer: Natural protein sequences have generally evolved to fold into stable functional structures. Because of this, many random amino-acid substitutions are expected to make folding less favorable. Experimental protein-stability datasets, including ProTherm-derived datasets, also contain more destabilizing than stabilizing mutations.

Answer: Yes. Since SAAFEC-SEQ requires only protein sequence information and does not require a 3D structure, it is suitable for large-scale or genome-scale studies of missense variants.

Answer: Please verify that:

• The submitted protein sequence is correct.
• The mutation position corresponds to the submitted sequence numbering.
• The wild-type amino acid in the mutation input matches the amino acid at that sequence position.
• The mutation is a valid single amino-acid substitution.
• The FASTA and mutation-list formats are correct.

Answer: No. SAAFEC-SEQ is intended for single-point missense mutations. Insertions, deletions, stop-gain mutations, frameshifts, and combined multi-residue mutations are outside the intended scope of the method.

Answer: No. SAAFEC-SEQ predicts the effect of a mutation on protein thermodynamic stability. Although stability changes can influence protein function and disease mechanisms, the predicted ΔΔG should not be interpreted as a direct measurement of protein activity, binding, expression, or pathogenicity.

Answer: Please contact us at delphi@g.clemson.edu. To help us reproduce the issue, please include the input sequence, mutation information, job id, error message, and, if available, a screenshot.