Zhang, Yong-Biao

Abstract Motivation Illumina DNA methylation arrays have evolved rapidly, expanding genomic coverage while introducing backward incompatibilities by removing many CpG sites present in earlier versions. These changes result in systematic missing values when integrating data across array generations and substantially limiting the reuse of legacy datasets. Results We developed a two-stage framework for imputing missing DNA methylation values. The procedure first imputes randomly missing values using standard imputation techniques and then addresses systematic missingness using multi-output machine learning models, including support vector regression, nearest-neighbor methods, random forest models, and deep neural networks. When evaluated on real datasets with up to fifty percent induced missingness, the proposed framework consistently outperformed conventional imputation approaches. It also accurately imputes the missing CpG sites between methylation arrays and reduced representation bisulfite sequencing data, enabling robust cross-platform data integration. Analyses of large brain tumor methylation datasets demonstrate that the method restores array-specific methylation patterns while preserving biological complexity. Importantly, imputing missing methylation sites significantly improves the performance of epigenetic age prediction models. Availability and implementation This tool is implemented in the Python package “ultra-impute,” freely available at https://github.com/liguowang/ultra-impute. A code snippet demonstrating the usage of the ultra-impute package is provided in a Jupyter Notebook (https://github.com/liguowang/ultra-impute/blob/master/doc/Tutorial.ipynb).

SUMMARY Artificial intelligence (AI)-based methods are increasingly critical in peptide drug discovery, but their applications are limited in a narrow scope such as antimicrobial peptides. For indications where peptide therapy inherently excels, such as metabolism, endocrinology, and tissue regeneration, AI-based pipeline for therapeutic peptide discovery is long awaited but still unmet. Here, we propose a deep learning-based pipeline, Deepeptide, capable of identifying therapeutic oligopeptides for various metabolism-related indications. Leveraging the intrinsical relationship of ‘disease indication — biological processes — molecular functions’, Deepeptide discovers oligopeptides with indication-ameliorating-related molecular functions as lead candidates for indication of interest. Deepeptide was applied in five representative indications of metabolism, endocrinology, and tissue regeneration: angiogenesis, lipid metabolism, osteogenesis, glucose metabolism, and anti-angiogenesis. Overall, 62% of the identified oligopeptide candidates demonstrated significant bioactivity in vitro, with most of them showing comparable potency to the first-line drugs. Notably, the heptapeptide AP7 exhibited angiogenic potency comparable to VEGF in excisional wound splinting mouse model by promoting cell migration rather than proliferation, and hexapeptide TP6 showed significant dual-efficacy against hyperlipidemia and obesity in high-fat diet mice by inhibiting lipid synthesis and regulating gut microbiota. These findings highlight the potential and generalizability of Deepeptide in therapeutic oligopeptide discovery for metabolic diseases.

AbstractCraniofacial microsomia (CFM, OMIM%164 210) is one of the most common congenital facial abnormalities worldwide, but it’s genetic risk factors and environmental threats are poorly investigated, as well as their interaction, making the diagnosis and prenatal screening of CFM impossible. We perform a comprehensive association study on the largest CFM cohort of 6074 samples. We identify 15 significant (P < 5 × 10−8) associated genomic loci (including eight previously reported) and decipher 107 candidates based on multi-omics data. Gene Ontology term enrichment found that these candidates are mainly enriched in neural crest cell (NCC) development and hypoxic environment. Single-cell RNA-seq data of mouse embryo demonstrate that nine of them show dramatic expression change during early cranial NCC development whose dysplasia is involved in pathogeny of CFM. Furthermore, we construct a well-performed CFM risk-predicting model based on polygenic risk score (PRS) method and estimate seven environmental risk factors that interacting with PRS. Single-nucleotide polymorphism-based PRS is significantly associated with CFM [P = 7.22 × 10−58, odds ratio = 3.15, 95% confidence interval (CI) 2.74–3.63], and the top fifth percentile has a 6.8-fold CFM risk comparing with the 10th percentile. Father’s smoking increases CFM risk as evidenced by interaction parameter of −0.324 (95% CI −0.578 to −0.070, P = 0.011) with PRS. In conclusion, the newly identified risk loci will significantly improve our understandings of genetics contribution to CFM. The risk prediction model is promising for CFM prediction, and father’s smoking is a key environmental risk factor for CFM through interacting with genetic factors.