Elucidating the molecular nexus between dental developmental anomalies and chronic inflammatory diseases: A computational systems biology analysis of genomic sequence homology using matrix factorization.

This entirely in silico, exploratory study investigates potential sequence-level links between periodontal disease, developmental dental anomalies such as molar-incisor hypomineralization (MIH), and selected systemic conditions including leukemias, amyotrophic lateral sclerosis, and Alzheimer's disease. To identify annotation-independent signatures, we applied Non-negative Matrix Tri-Factorization (NMTF) to k-mer/tf-idf representations of disease-associated gene and protein sequences compiled from OMIM, GeneCards, and PubMed for eleven curated disease groups; similarity values denote the proportion of the total between-cluster interaction (S-matrix) captured by a cluster. To guard against trivial explanations, we applied a composition-preserving sequence-shuffling test of the similarity magnitude and a degree-preserving (curveball) permutation test of the disease co-clustering that accounts for the dependence introduced by isoform records and shared genes, repeating the latter across a 1890-configuration parameter grid. At the gene level the dominant cluster was governed by the largest disease group (MIH) rather than by disease identity, whereas at the protein level periodontitis, dental anomalies, and leukemia subtypes co-clustered. The magnitude of the shares did not exceed a composition-preserving null (gene p=0.66; protein p=0.082), and under the degree-preserving null co-clustering was not significant at the gene level (p=0.36) but was significant, modest, and grid-robust at the protein level (p=0.033); canonical k-mer and coding-sequence controls confirmed that the gene-level weakness is intrinsic to nucleotide k-mers. These results delineate the protein level as the informative scale for annotation-independent cross-disease screening and identify the dental-leukemia axis as the candidate relationship for targeted follow-up. The findings are hypothesis-generating: the disease-associated gene sets were compiled with a non-systematic, literature-guided procedure, magnitudes do not exceed composition baselines, neither experimental or clinical validation nor quantitative benchmarking against alternative computational frameworks was performed, and any translational implication requires future validation.