Spotlight · Journal article · 2025
A comprehensive targeted panel of 295 genes in multiple myeloma (BIO-DGI)
Attention-based graph learning across SNVs, CNVs and SVs to find disease-initiating biomarkers.
Computational genomics · Columbia University
Reading genomes to explain disease, not just predict it.
I build ancestry-aware statistical and interpretable AI methods that connect genetic variation to genes, pathways and networks in Alzheimer's disease and other complex disorders.
Track 1 · Journey
My career, drawn as a chromosome.
Each band is a chapter, positioned by year. Tap a band to read its locus: where it happened, what I built, and what it taught me.
Track 2 · Work
A Manhattan plot of everything I've published and presented.
Each point is a patent, paper, preprint, abstract, talk or tool. Left to right is the date; height is the type of output, from software at the bottom to patents at the top. Everything above the dashed line is peer-reviewed or granted, the way a GWAS peak clears genome-wide significance. Bigger journal points mean a higher impact factor.
5Journal articles
12.9Highest impact factor · Blood Cancer Journal
28.9Cumulative impact factor
4Oral talks
Show all 17 as a list
- 2026Maternal-offspring haplotype transmission in ADSP dyadsOral talkNIA, NIH · Rockville
- 2026Maternal-offspring haplotype transmission reveals AD-relevant signals in ADSP dyadsOral talkAD/PD 2026 · Copenhagen
- 2025Gene-based approaches integrating maternal and child genetics in FASDOral talkUCLouvain · Brussels
- 2025Gene-Based Genome-Wide Approaches Integrating Maternal And Child Genetic Data Identify Novel Risk- Modifying Genetic Loci For Fetal Alcohol Spectrum DisordersConference abstractESBRA 2025
- 2025
- 2025Colocalization of eQTL and GWAS in a multiethnic brain bankConference abstractAAIC 2025
- 2025Journal articleComputers in Biology and MedicineIF 7.0
- 2025Fine mapping of regions of homozygosity in Alzheimer's disease: Peruvian and Mexican populationsOral talkAD/PD 2025 · Vienna
- 2025SoftwareR / Shiny
- 2024Gene × prenatal alcohol exposure effects on growth and cognition in mother-child dyadsConference abstractASHG 2024
- 2023BDL-SP: a bio-inspired DL model for altered signalling pathways in multiple myelomaJournal articleAmerican Journal of Cancer ResearchIF 3.1
- 2022Journal articleFrontiers in BioinformaticsIF 2.8
- 2022SoftwareGitHub · Apache-2.0
- 2022Characterizing the mutational landscape of MM and its precursor MGUSJournal articleAmerican Journal of Cancer ResearchIF 3.1
- 2021Conference abstractClinical Lymphoma Myeloma and Leukemia
- 2021
- 2020Journal articleBlood Cancer JournalIF 12.9
Track 3 · Thought process
How I move from a variant to an explanation.
Most of my projects follow the same path. The question at each step matters more than the method.
Variant
What changed?
Whole-genome and exome variation, runs of homozygosity, ancestry.
Gene
Which gene carries it?
Gene-level association, eQTL mapping, fine-mapping, colocalization.
Pathway
What process breaks?
Signalling pathways and biologically informed deep learning.
Network
Who else is affected?
Gene-gene interaction graphs and attention-based graph learning.
Explanation
Can I test it?
A short list of genes and mechanisms someone can take to the bench.
Explain why, not just what.
A model that predicts well but cannot say which genes matter has not finished the job.
Ancestry is part of the answer.
Signals found in one population often look different in another. Diverse cohorts are the point, not a robustness check.
If others can't run it, it isn't done.
miRPipe, miRSim, BRAINscape and SAGA exist so the next lab starts further ahead.
Track 4 · Ambitions
Open hypotheses I want to spend the next decade on.
Written like a lab notebook so they can change as the evidence does.
Active
Ancestry-specific regulation explains part of Alzheimer's risk that European-only studies miss.
Testing with eQTL and GWAS colocalization in multiethnic brain banks.
Active
Inheritance patterns between mother and child carry disease signals that case-control designs overlook.
Haplotype transmission in ADSP dyads and gene × environment effects in birth cohorts.
Next
Biologically informed graph models can be interpretable enough to generate lab-testable hypotheses.
Extending the BIO-DGI approach from blood cancer to neurodegeneration.
Long term
An independent lab building open, ancestry-aware multi-omics resources.
Methods and tools that make discovery generalize across populations.
Track 5 · Notebook
Papers explained, methods, and new ideas.
Longer notes on the work above, written for people outside my subfield.
Paper explained · August 19, 2026BDL-SP: Using Bio-Inspired Deep Learning to Identify Altered Signaling Pathways in Multiple MyelomaA beginner-friendly and technical guide to BDL-SP, a graph-based explainable deep-learning framework for distinguishing MGUS from multiple myeloma and identifying altered genes and pathways from whole-exome sequencing data.Paper explained · August 19, 2026From Genomic Complexity to a 295-Gene Panel: How BIO-DGI Studies Multiple Myeloma ProgressionA beginner-friendly and technical guide to BIO-DGI, an explainable graph-learning framework used to derive a 295-gene research panel for studying the transition from MGUS to multiple myeloma.Method · October 28, 2025miRPipe: Beginner and Advanced User GuideAn illustrated guide to the purpose, workflow, use, interpretation, applications, benchmarking, and limitations of miRPipe.
Track 6 · Open tools