Nephrologist & Internal Medicine Specialist · Precision Medicine and Bioinformatics · Clinical Data Science · Real-World Evidence

Practicing nephrologist in the Dominican Republic. I see the patients and I analyse the data — clinical questions come from the consultation room and the dialysis unit, and the answers are built with reproducible transcriptomic, proteomic and real-world clinical data pipelines.

📍 Santo Domingo / Baní, Dominican Republic · 🔗 LinkedIn

Clinical Nephrology → Real-World Clinical Data → Bioinformatics → Multi-Omics → Precision Medicine

The bridge is the point. Molecular datasets are interpreted with a nephrologist's understanding of the disease; clinical datasets are analysed with a bioinformatician's discipline about reproducibility, cohort definition and what a result can and cannot support.

Question · Are molecular signals observed in kidney tissue preserved at the protein level in urine? Data · Kidney tissue RNA-seq (GSE142025: control / early DKD / advanced DKD) + published urinary proteomics (~239 samples, DKD stage 3 vs 4). Unpaired, cross-compartment. Methods · limma differential expression (adj. p < 0.05, |logFC| ≥ 1), gene-symbol harmonisation, concordance classification. Result · 1,743 significant genes and 555 detected proteins, 81 overlapping. Concordance: 42 down–down, 9 up–up, 30 discordant. Dominant signal is metabolic/tubular decline (SORD, GSTA1/2, ALDH1L1, ASS1, MME) with complement-linked inflammatory activation (C3, CFH) — not fibrosis alone. Relevance · Supports the feasibility of tracking tissue-level DKD biology through non-invasive urinary markers, and shows where that translation breaks down. Stack · R (limma), Python · Status · Research project, public + published data; manuscript in preparation → dkd-multiomics-fibrosis-metabolism-signature

Question · Is intradialytic hypotension driven by the dialysis session, or by the patient? Data · 394 hemodialysis sessions from 52 patients, extracted directly from Nikkiso DBB-06 machines at a Dominican dialysis unit. Real-world, longitudinal, machine-derived. Methods · Feature engineering (ΔMAP, maximum systolic drop, UFR, IDWG), descriptive comparison by IDH status, a simple clinical risk score, and patient-level K-means clustering (k = 3). Result · IDH rate 43.15% (170 events). Classical session-level predictors (UFR, IDWG, haemoglobin) discriminated poorly. Three hemodynamic phenotypes emerged; the most unstable one was not the highest-UFR group. Relevance · Argues for phenotype-based risk stratification and individualised ultrafiltration rather than session-parameter thresholds. Stack · Python (pandas, scikit-learn) · Status · Retrospective observational analysis; exploratory ML model presented as poster and oral conference at the XIV Congreso Dominicano de Nefrología / VI Encuentro Mayo Clinic (2025); manuscript in preparation → hemodialysis-intradialytic-hypotension-risk-analysis

Question · What is the CKD burden detected by primary-care screening in a low-resource setting, and which social barriers shape it? Data · UNAPS primary-care screening cohort, n = 400, Peravia province; nested sociodemographic subcohort, n = 50 linked patients. Methods · Reproducible Python pipeline: data audit → cleaning → derived-ID linkage → quality-control flagging of discordant fields → descriptive epidemiology → sociodemographic analysis. Result · Substantial renal-risk burden with hypertension and diabetes as dominant drivers; the linked subcohort surfaces education, insurance coverage and economic barriers that clinical variables alone do not capture. Linkage inconsistencies were flagged rather than silently harmonised. Relevance · Directly usable for prevention policy and health-system planning in underserved settings. Stack · Python (pandas, matplotlib) · Status · Cross-sectional descriptive study; manuscript in preparation → ckd-primary-care-dominican-republic

Question · Can proteomic profiling identify renal-cancer subgroups with distinct survival, and is the signal reproduced at RNA level? Data · TCGA-KIRC, 475 patients with matched RNA-seq and RPPA. Methods · 11-script reproducible R workflow: acquisition → cohort matching → unsupervised clustering → survival comparison → cross-layer marker discovery. Result · Two proteomic clusters; the smaller subgroup (n = 86) showed a higher event rate (~39% vs ~34%) and concordant protein/RNA signatures of proliferative signalling, DNA-repair activation and metabolic dysregulation. Relevance · Demonstrates cross-layer multi-omics integration with survival endpoints; hypothesis-generating, not a prognostic classifier. Stack · R (survival, clustering, differential analysis) · Status · Public-data reanalysis; no external validation → tcga-kirc-multiomics-survival-signature

Question · Which cell populations and cell states carry the injury signal in human kidney tissue? Data · GSE131685, human kidney scRNA-seq. Methods · Seurat v5 pipeline — QC (nFeature_RNA 200–6,000; mitochondrial ≤ 15%), normalisation, clustering, manual marker-based annotation. Result · Resolved nephron segments (proximal tubule, distal tubule, collecting duct) alongside T/NK, B and myeloid populations, with transcriptional programmes consistent with oxidative stress and epithelial injury across multiple compartments. Relevance · Kidney injury reads as a multi-compartment process, not a single-cell-type event — the framing that precision nephrology depends on. Stack · R (Seurat v5, dplyr, ggplot2) · Status · Public-data reanalysis; manual annotation, no trajectory analysis → human-kidney-singlecell-injury-transcriptomic-analysis

hemodialysis-survival-catheter-vs-fistula-ml — XGBoost + SHAP model for 1-year mortality in hemodialysis, examining vascular access, inflammation and nutritional status. The dataset is synthetic (n = 2,500), clinically grounded but not a real registry. Held-out test ROC-AUC 0.758; cross-validated ROC-AUC ≈ 0.63 (± 0.04), i.e. moderate and unstable. Built to demonstrate modelling, interpretability and clinical reasoning — not a validated or deployable clinical tool.

Chronic kidney disease · Diabetic kidney disease · Glomerular disease (FSGS, lupus nephritis) · Hemodialysis outcomes and risk stratification · Kidney precision medicine · Biomarker discovery · Multi-omics integration · Transcriptomics and single-cell analysis · Real-world evidence · Clinical epidemiology in low-resource settings

R — limma, DESeq2, Seurat v5, clusterProfiler, survival, Bioconductor, renv Python — pandas, scikit-learn, XGBoost, SHAP, matplotlib Data & reporting — SQL, Power BI Practice — reproducible pipelines, scripted end-to-end workflows, documented QC and linkage decisions, version control

• El futuro de la Nefrología Dominicana — exploratory machine-learning model for intradialytic hypotension. Poster and oral conference, XIV Congreso Dominicano de Nefrología / VI Encuentro Mayo Clinic, 2025.
• Estrategias de Prevención y Manejo de la Enfermedad Renal Crónica en la Población Rural — invited lecture, 2021.

Manuscripts in preparation: DKD multi-omics; intradialytic hypotension phenotyping; CKD in Peravia primary care. No claim of acceptance or publication is made for these.

Kidney precision medicine for Latin American and Caribbean populations; non-invasive biomarkers in diabetic kidney disease; phenotype-based risk stratification in dialysis; making real-world clinical data from low-resource health systems usable for research.

MD, Universidad Autónoma de Santo Domingo · Internal Medicine (2016) and Nephrology (2019), UASD / Hospital Docente Padre Billini · MSc in Bioinformatics (Máster Universitario en Bioinformática), Universidad Alfonso X el Sabio, Spain — studies completed July 2026

Member, Scientific and Research Committee — Sociedad Dominicana de Nefrología (SODONEF), 2026 Board · Research Committee Board Member, Hospital Nuestra Señora de Regla

Disclaimer — Every repository here is research or methodological work. None of it is a validated clinical decision-support tool, none has regulatory clearance, and none should be used for patient-level decisions. Datasets are public, published, de-identified, or synthetic; raw identifiable clinical data are not shared.

📩 Open to collaboration and to roles in precision medicine, translational and clinical research, clinical data science and real-world evidence — LinkedIn · ORCID 0009-0009-7503-222X