Presentations

In April 2026, my team won 1st place at the University of Warwick Bioimage 2026 Hackathon, a three-day event funded by Warwick Spotlight. The challenge was to predict which of twelve antibiotics had been used to treat E. coli cells based purely on cellular morphology, captured across four fluorescence channels (DNA, permeability, cell wall, and membrane).

We built various classifier trained on ~80 handcrafted morphological features extracted from over 24,000 segmented cells across 37 wells. UMAP visualization revealed that cells cluster by morphological outcome rather than molecular mechanism: drugs targeting different pathways (Ampicillin vs Naladixic acid) cluster together when they produce the same phenotype (filamentation), while mechanistically related drugs separate when their cellular consequences differ.

Why this matters: traditional antibiotic discovery is slow because identifying a new compound's mechanism of action typically requires expensive biochemical assays. Morphology-based profiling, using only microscopy data, can dramatically accelerate this process by classifying candidate compounds into mechanism classes within hours. As antibiotic resistance becomes one of the defining public health challenges of the century, scalable phenotypic screening pipelines like this one help researchers triage thousands of candidate compounds and prioritise those most likely to act through novel mechanisms.