// bioinformatics · ML · population genetics
Jikaël Ntoko
M.Sc. Molecular & Life Sciences — Bioinformatics · UNIL
"Where rigorous data science meets real biological questions."
scroll
// bioinformatics · ML · population genetics
M.Sc. Molecular & Life Sciences — Bioinformatics · UNIL
"Where rigorous data science meets real biological questions."
I am a computational biologist finishing my Master's in Molecular & Life Sciences — Bioinformatics at the University of Lausanne. My work sits at the intersection of population genetics, immunoinformatics, and machine learning, with a strong emphasis on building rigorous, reproducible pipelines.
During my thesis in the Goudet Group, I built large-scale Python/R simulation frameworks using SLiM 5.0 to investigate bias in the QST–FST evolutionary framework — running 500 replicates per condition across thousands of HPC jobs on SLURM. Before that, I spent six months in the Gfeller Group benchmarking PyTorch architectures for MixTCRpred, a T-cell receptor–epitope interaction predictor.
I thrive where statistical rigor and software engineering discipline meet real biological questions — whether that's detecting selection from genomic data, optimizing deep learning models, or building production-grade pipelines with CI/CD from day one.
// project_01
Deep learning framework to distinguish local adaptation from genetic drift — directly targeting the failure modes of the QST–FST framework identified in my thesis. Implements a CNN operating over the spatial axis of allele frequency matrices to detect selection signals even under isolation-by-distance scenarios where classical tests systematically fail.
// project_02
Production-grade convolutional Variational Autoencoder in PyTorch for automated, unsupervised cell phenotype classification. Features automated t-SNE visualisation of the latent space, full CI/CD pipeline via GitHub Actions and Poetry, and unit tests enforced on every commit — demonstrating software engineering discipline well beyond typical academic projects.
// project_03
Large-scale Python/R simulation framework using SLiM 5.0 to model polygenic trait architectures (additive, dominant, recessive, epistatic) across 1–1,000 QTLs in two population structures. 500 replicates per condition, parallelised on SLURM HPC. Showed FPR drops from ~30% to ≤5% under additive polygenic models (≥5 loci), and identified conditions where QST–FST structurally breaks down.
Bias and robustness of the QST–FST framework under polygenic and non-additive trait architectures Master's Thesis
University of Lausanne · Department of Ecology and Evolution · Sep 2025
Benchmarking CDR input encoding strategies for MixTCRpred: full CDR1/2 sequences are necessary for peak T-cell receptor–epitope interaction prediction Research Internship
Internal Report · Gfeller Lab, UNIL · Jun 2024 · AUC 0.88 on 19 epitopes, 5-fold cross-validation
DeepLocalAdaptation: a CNN-based approach to distinguish local adaptation from genetic drift beyond the limits of QST–FST
Work in Progress · github.com/JikaelN/DeepLocalAdaptation
I'm actively looking for positions in computational biology, bioinformatics, or ML-for-life-sciences — in Switzerland or remotely. Whether it's a research role, internship, or industry position, I'd love to hear from you.
jikael@ntoko.ch