The FASTA sequence of the human ARSB gene was obtained from the UniProt database (UniProt ID: P15848). A database of missense mutation SNPs was created using information gathered through a PubMed literature review and databases such as dbSNP, HGMD, ClinVar, and Ensemble. The list was cleared of the redundant nsSNPs. The Protein Data Bank (PDB ID: 1FSU) provided the crystal structure of human ARSB. The remaining mutations were obtained from the Ensembl databases, and all mutation types, including missense, non-synonymous, and synonymous mutations, were included (Dyer et al., 2025).

Sorting Intolerant From Tolerant (SIFT: http://sift.jcvi.org/) was used to predict the functional impact of amino acid substitutions based on sequence homology and evolutionary conservation (Sim et al., 2012). The underlying principle is that functionally important residues are conserved across species; therefore, substitutions at highly conserved positions are more likely to be deleterious. SIFT assigns a tolerance score ranging from 0 to 1, where scores $\le 0.05$ indicate damaging (intolerant) substitutions and scores $>0.05$ suggest tolerated variants. In addition to missense variants, SIFT can also evaluate certain 3-nucleotide indels that result in amino acid insertions or deletions. These predictions are based on conservation patterns derived from multiple sequence alignments.

PolyPhen-2. PolyPhen-2 (http://genetics.bwh.harvard.edu/pph2/) used to predict the potential structural and functional impact of amino acid substitutions using sequence-based and structure-based features. It incorporates evolutionary conservation, physicochemical differences between residues, and structural parameters such as solvent accessibility and proximity to functional domains (Adzhubei et al., 2013). The tool calculates the difference in Position-Specific Independent Count (PSIC) scores between wild-type and mutant residues. Variants are classified as benign, possibly damaging, or probably damaging, depending on the PSIC score difference and predictive confidence.

Mutation Assessor. Mutation Assessor (http://mutationassessor.org/) evaluates the functional impact of amino acid substitutions based on evolutionary conservation patterns within protein families and subfamilies (Sarachakov et al., 2025). It distinguishes between general conservation across homologous proteins and functional specificity within subfamilies. Each variant is assigned a Functional Impact (FI) score and categorized as neutral, low, medium, or high impact. Variants with FI scores $>2.0$ are generally considered functionally deleterious.

FATHMM. Functional Analysis Through Hidden Markov Models (FATHMM) predicts the pathogenicity of missense mutations using Hidden Markov Models (HMMs) and conservation-based weighting schemes (Shihab, 2021). It can incorporate disease-specific or cancer-specific models. Variants are classified as tolerated or deleterious based on a threshold score, with lower scores indicating a higher likelihood of pathogenicity.

MAESTROweb. MAESTROweb (https://pbwww.che.sbg.ac.at/maestro/web) predicts mutation-induced changes in protein stability using machine-learning methods trained on experimental thermodynamic datasets. It calculates the change in Gibbs free energy ($\Delta \Delta G$) between wild-type and mutant proteins (Laimer et al., 2016). A negative $\Delta \Delta G$ value indicates destabilization of protein structure, whereas a positive value suggests stabilization. The tool also identifies mutation "hotspots" by scanning multiple residues.

mCSM. mCSM predicts the impact of mutations on protein stability and macromolecular interactions using graph-based structural signatures derived from atomic distance patterns. It estimates $\Delta \Delta G$ values, where negative scores indicate destabilizing mutations (Pires et al., 2014). Variants with significantly negative $\Delta \Delta G$ values are considered likely to disrupt protein stability and function.

DynaMut2. DynaMut2 (http://biosig.unimelb.edu.au/dynamut/) integrates Normal Mode Analysis (NMA) with graph-based signatures to evaluate mutation-induced changes in protein stability and dynamics (Rodrigues et al., 2021). It predicts both $\Delta \Delta G$ and changes in vibrational entropy ($\Delta \Delta S$), allowing assessment of alterations in structural flexibility. This tool is particularly useful for understanding how mutations affect conformational dynamics, well to thermodynamic stability.

PremPS. PremPS (https://lilab.jysw.suda.edu.cn/research/PremPS/) predicts the impact of missense mutations on protein stability using a balanced dataset of stabilizing and destabilizing mutations (Chen et al., 2020). It integrates structural and evolutionary features to estimate $\Delta \Delta G$ values and classify variants as stabilizing or destabilizing.

MutPred2. MutPred2 (http://mutpred.mutdb.org) is a machine learning-based tool that predicts the pathogenicity of amino acid substitutions and provides mechanistic insights into possible molecular alterations (Pejaver et al., 2020). It assigns a pathogenicity probability score ranging from 0 to 1. Variants with scores $>0.5$ (commonly $>0.589$ for high confidence) are considered likely pathogenic. Additionally, it predicts alterations in secondary structure, post-translational modification sites, catalytic residues, and protein-protein interactions.

SNPs&GO. SNPs&GO integrates sequence features and Gene Ontology (GO) annotations using a Support Vector Machine (SVM) classifier to distinguish disease-associated variants from neutral polymorphisms. The inclusion of functional annotations improves predictive accuracy (Capriotti et al., 2013).

PhD-SNP. PhD-SNP (https://snps.biofold.org/phd-snp/phd-snp.html) is an SVM-based predictor that classifies missense mutations as disease-associated or neutral using sequence and evolutionary profile information. Variants with scores $>0.5$ are predicted to be disease-causing (Calabrese et al., 2008).

ConSurf. ConSurf (https://consurf.tau.ac.il/) evaluates evolutionary conservation of amino acid residues using multiple sequence alignment and phylogenetic analysis (Ashkenazy et al., 2016). Conservation scores range from 1 (variable) to 9 (highly conserved). Highly conserved exposed residues are often functionally important, whereas conserved buried residues are typically structural (architectural). Disease-associated mutations frequently occur at highly conserved positions.

SODA. SODA (Solubility based on Disorder and Aggregation) predicts the impact of mutations on protein solubility, secondary structure, and aggregation propensity. It helps identify variants that may promote protein misfolding or aggregation, which is relevant in disorders associated with lysosomal dysfunction (Paladin et al., 2017).

Arpeggio. Arpeggio analyzes interatomic interactions within protein structures by categorizing them into hydrogen bonds, van der Waals contacts, hydrophobic interactions, and ionic interactions. It accepts PDB structures and generates quantitative interaction profiles, enabling comparison between wild-type and mutant proteins to assess structural disruption (Jubb et al., 2017).

Sequence-based evaluation was performed on all nsSNPs utilising five online tools: SIFT, PolyPhen2, PROVEAN, Mutation Assessor, FATHMM Using the structure-based approach, 141 nsSNPs located in the ARSB gene were analysed (Table S1) using PremPs, MAESTROweb, DynaMut2, and mCSM. As the sole region identified through experimentation and available in the PDB database, only the highly probable nsSNPs have been selected for additional investigation. Using the PMut web server and MutPred2, high-probability nsSNPs' disease phenotypes were identified. We utilised FATHMM, SIFT, PolyPhen2, and SNPs&GO. Based on protein physical characteristics, SIFT classifies variants as either intolerant or tolerated; higher scores indicate a higher likelihood of toxicity. The study of 1224 single-point amino acid changes in ARSB using the sequence-based approach produced predictions from SIFT, PolyPhen2, Mutation Assessor, and FATHMM. These tools, particularly, highlighted the dangerous substitutions 302, 264, 238, and 167 (Figure 2). By combining these sequence-based predictive techniques, we were able to examine the possible molecular effects of mutations on functioning.

We used four different structure-based prediction technologies: DynaMut2, MAESTROweb, PremPS, and mCSM (Figure 2). By computing folding free energy, these tools assess the stability of variations by analysing atomic coordinates from the PDB file of the wild-type protein. Many of these tools use a machine-learning approach, combining different biophysics-based methodologies to predict the impact of variations on protein stabilisation. Folded (Gf) and unfolded (Gu) forms are computed in thermodynamics using the formula $\Delta G=G_u-G_f$. $\Delta \Delta G=G_m-G_w$, wherein $G_m$ is the energy of the protein that has been altered, and $G_w$ is the energy of the wild-type protein, is used to assess the change in protein stability and the landscape of energy. A variation that stabilises a protein is indicated by a negative $\Delta \Delta G$ value, whereas a mutation that destabilises the protein is suggested by a positive $\Delta \Delta G$ score.

We evaluated disease characteristics linked to mutations using PhD-SNP, SNAP & GO, and MutPred2 approaches. These tools classify variants based to potential disease associations and assign pathogenicity levels (Figure 4). Among the 57 high-confidence variations identified through a sequence- and structure-based evaluation, PMut and MutPred predicted 44 as pathogenic. Only 44 mutations, A237D, C91R, E323K, E483D, G149R, G302R, G324V, G56D, G64R, H393P, I296N, I67N, K145E, L129P, L236P, L360P, L498P, L51P, L72P, L72R, L82P, L82R, L90P, L98P, L98R, P93R, R315P, R315Q, R327G, R327Q, R388T, T92K, V277G, V80G, W146R, W146S, W353R, W438G, Y138C, Y175D, Y210C, Y266S, Y86C, Y86N) out of 57 highly probable nsSNPs were shown to be harmful using the disease phenotype assessment algorithm (Table 3).

The ConSurf tool was utilised to analyse the human ARSB genome structure and assess residue conservation (Ashkenazy et al., 2016). According to the ConSurf study, out of 44 final mutations, these 12 (A237D, G56D, G64R, L51P, I67N, L236P, W353R, V277G, T92K, R315Q, R315P, H393P) can be highly disease-causing mutations, because they belonged in high conserved areas and mutations in these regions (Figure 5).

The aggregation, illness, helix, and strand propensities resulting from alterations are computed using SODA48. Out of 12 nsSNPs, 2 (A237D, W353R) were shown (Table 1) to lower the protein’s solubility, and 9 of them raised it, out of the 11 mutations revealed using illness phenotype prediction. The solubility of a protein significantly impacts its function. Insoluble proteins tend to congregate, which can lead to illnesses including Parkinson’s, amyloidosis, and Alzheimer’s. By combining solubility data with structural and sequence-based properties, SODA provides insights into protein regions susceptible to aggregation and suggests modifications that may improve solubility (Aggidis et al., 2024).

Treatment plans to mitigate the effects of such harmful mutations can be developed using the knowledge gained from this study (Israil et al., 2025). Creating medications or tiny compounds that stabilise the peptide, improve its ability to dissolve, or stop it from aggregating are some possible strategies. Additionally, by understanding the structural alterations caused by these mutations, tailored therapies that restore lost or normal protein function may be developed.

In the final analysis, our thorough examination of nsSNPs in the ARSB gene not only advances our understanding of the genetic underpinnings of associated illnesses but also lays the groundwork for developing effective treatments. The integration of structural, solubility, and sequencing research offers a comprehensive knowledge of the effects of pathogenic mutations. Two of the discovered mutations are located in highly conserved regions that may affect the gene's function, according to our ConSurf analysis. Using the Arpeggio web server, we further evaluated the aggregation properties and conducted additional structural analyses (Figure 7). We discovered two changes that reduce protein solubility, suggesting a possible role in ARSB aggregation and the ensuing illness. The identification of mutations with the greatest potential to contribute to disease pathophysiology was made possible by sequence-based analysis. Simultaneously, the predictions based on structure reinforced our belief that mutations could be crucial to the onset and progression of the disease.

The wild-type ARSB protein, along with a few selected mutant variants, was subjected to molecular dynamics (MD) simulations to confirm the structural consequences of high-confidence deleterious substitutions. MD simulations provide dynamic insights into the stability, flexibility, and conformational behavior of proteins under physiologically relevant conditions, in contrast to static protein structure investigations.