2026-10-14T23:21:58Z https://www.scidb.cn/oai

10.57760/sciencedb.44397 2026-07-31T16:04:31Z

2026-07-31 Structure-aware prediction of functional non-synonymous variants in a maize EMS mutant library doi:10.57760/sciencedb.44397 en Prediction of mutational effects of non-synonymous mutations Predicted protein structures (PDB and mmCIF formats) of the maize reference proteome (UniProt: UP000007305) were obtained from the AlphaFold Protein Structure Database (https://alphafold.com/download). Genome annotation, protein sequences, and coding sequences (CDS) of the B73 genome were downloaded from MaizeGDB (https://download.maizegdb.org/Zm-B73-REFERENCE-GRAMENE-4.0/). Custom Python scripts integrating the Bio.Seq module and the pandas library were used to ensure correspondence between predicted structures and protein sequences, particularly for sequences containing non-synonymous mutations annotated by SnpEff (Cingolani et al., 2012). Briefly, protein sequences were extracted from AlphaFold PDB files using Bio.Seq and aligned to the reference protein sequence file from MaizeGDB (Zm-B73-REFERENCE-GRAMENE-4.0_Zm00001d.2.protein.fa). Entries with identical alignment results were considered matched protein sequences and predicted structures. Annotation information and mutation data were then integrated using the pandas library to construct datasets of protein sequences harboring non-synonymous mutations. In addition, we analyzed codon composition and mutational patterns based on the protein sequence datasets described above. First, we summarized the total number of occurrences of each amino acid and their corresponding codons to assess amino acid–codon correspondence. Next, we simulated GC-to-AT substitutions for all codons containing G or C and calculated the weighted proportion of each possible codon change using codon frequency as a weighting factor to estimate the expected occurrence probability of GC-to-AT mutations. For mutations resulting in non-synonymous substitutions, we further calculated the probability distribution of each mutation type among all non-synonymous changes. Finally, we utilized the Python scikit-learn library to perform linear regression analysis, comparing the estimated proportions of non-synonymous mutation types with the observed proportions derived from EMS mutagenesis and calculating the residuals between predicted and observed values. To predict the mutational effects of non-synonymous substitutions, we used SaProt (Su et al., 2024), with source code and model weights obtained from GitHub (https://github.com/westlake-repl/SaProt). SaProt is a structure-aware protein language model trained on large-scale protein squence-structure data, in which residue tokens are jointly encoded with structure-derived tokens. For evolutionary conservation analysis, residule-level conservation scores were independently estimated using the ConSurf with default parameters (Ben Chorin et al., 2020). Briefly, homologous sequences were automatically retrieved from the UniProt, followed by redundancy reduction and representative sampling. Multiple sequence alignment was constructed internally by the server, and residue-specific conservation scores were estimated using a phylogeny-aware approach based on an automatically selected substitution model (LG) within the Rate4Site framework. Because the structural information used by SaProt is derived from AlphaFold2 predictions, regions with low predicted Local Distance Difference Test (pLDDT) scores were masked to improve prediction accuracy when converting protein structures into structure-aware sequences. The absolute value of the predicted mutational effects reflects the magnitude of the effect. Mutation effect (ME) is defined as the log-likelihood ratio comparing the wild-type amino acid to the mutant amino acid at the mutated residue position. When the probability of the mutant (mut) amino acid is higher than that of the wild-type (wt), ME takes a negative value; conversely, when the wild-type amino acid is more probable under the model, ME is positive (Su et al., 2024). A positive ME indicates that the mutant residue is more probable under the model given the sequence and structural context, whereas a negative ME indicates the opposite. From a biological perspective and in the context of SaProt-derived fitness prediction, this metric reflects the evolutionary fitness alteration of amino acid substitutions. Positive ME values correspond to elevated evolutionary fitness, representing well-tolerated or even strong-predicted-effect mutations, while negative ME values denote reduced evolutionary fitness, which commonly exerts moderate predicted impacts on protein function and structural stability. Notably, the majority of amino acid replacement mutations yield negative ME values and lead to decreased protein fitness. In addition, GPSite (Yuan et al., 2024) was used to predict the likelihood of amino acid interactions with different biomolecules, such as DNA, RNA, peptides, and ATP. GPSite output scores ranged from 0 to 1, with higher scores indicating a higher predicted binding likelihood. SaProt; GPSite; Mutational Effect; biomolecules Xuerong Yang PUBLIC https://creativecommons.org/licenses/by-nc/4.0/ dataset Science Data Bank