Projects

Corbeil Lab, Université Laval · Summer 2025 – Present

ML lead on a closed-loop antimicrobial peptide discovery system paired with a wet-lab collaborator running a 10K-sample high-throughput potency assay. A fine-tuned transformer with a custom regression head predicts log-fold-change activity across an 870K-peptide database; top and most-uncertain predictions drive the next testing batch (~25% hit rate expected on the first round, validation ongoing). Now exploring stronger pretraining inductive bias and generative models (diffusion, GFlowNets) for direct candidate design. Sponsored by NSERC and FRQNT.

Scientific Reports (Nature), 2026 · First author

Progress in computational AMP discovery has been hard to quantify because of inconsistent datasets and evaluation protocols. QMAP is a domain-specific, homology-aware benchmark for predicting antimicrobial potency (MIC) and hemolytic toxicity (HC50), with predefined test sets that enforce strict sequence-homology constraints between training and test data so performance reflects true generalization rather than overfitting. Reassessing existing MIC models and establishing new baselines on QMAP reveals limited real progress in the field over six years, poor performance on high-potency MIC regression, and low predictability for hemolytic activity.

Technical report, targeting ICLR 2027 · First author

Random splitting can yield non-independent train--test subsets when a dataset contains related samples, as is common in certain applications such as biochemical studies. This leads to overly optimistic generalization estimates. Here, we introduce ReLaG, a modality-agnostic framework that models sample relatedness through a hierarchical latent-variable process and infers groups of related samples using proximity graphs and community detection to produce independent train--test subsets. Across molecular and protein datasets, ReLaG matches existing relation-aware methods while scaling substantially better, enabling splits at previously impractical dataset sizes. We further introduce a label-free procedure that adapts the splitting resolution to production data, aligning evaluation with the intended deployment setting. ReLaG's inferred groups provide a cheap estimate of effective dataset size, enabling diversity-aware dataset scaling. ReLaG is open source and can be installed with: pip install relag.

ASMS Conference, San Diego, CA, 2026 · First author (poster)

A maximum-likelihood decomposition (MLD) algorithm for identifying bacterial species in polymicrobial MALDI-TOF mixtures, where standard identification workflows break down. Reference profiles are built from Gaussian-fitted, DBSCAN-clustered peaks from pure-culture spectra, and mixtures are decomposed by scoring how well each reference profile explains the observed spectrum via a log-likelihood ratio. On synthetic mixtures the method correctly identifies all components (likelihood ratios 1.5–1.7); experimental validation is ongoing, with mass-shift calibration (lock-mass) identified as the next step to improve robustness.

NORLAB, Université Laval · Fall 2025 – Present (part-time)

Visual SLAM robust to high-dynamic-range scenes where standard auto-exposure saturates and corrupts position/map estimates. After a deep-learning exposure predictor proved unreliable, switched to capturing multiple fixed-exposure images per frame and added custom factors so exposure changes no longer destabilize the estimate. Grew from a course project into an ongoing collaboration with NORLAB.

AI Club, Université Laval · 2024 – Present · Team Lead

Leading a student team building autonomous FPV drones that compete in a laser tag game, covering hardware, perception, control, and the autonomy stack.

Corbeil Lab, Université Laval · Summer 2024

Research internship (NSERC/FRQNT-funded) building a deep learning AMP potency predictor — later found to be affected by the data-leakage issue that motivated QMAP.

FLC Lab (Flavie Lavoie-Cardinal), Université Laval · Summer 2023

Applied active learning and anomaly detection to identify abnormal synapses in super-resolution microscopy images as potential early precursors of Parkinson's disease.