Publications

A 9-class differential diagnosis system for severe central nervous system infections, with primary amoebic meningoencephalitis (PAM) caused by Naegleria fowleri as the highest-stakes target class. Built in PyTorch on the RigoBERTa Clinical encoder, with a 5-model deep ensemble, dual-gate out-of-distribution detection (Mahalanobis plus logit-energy), and Mondrian per-class conformal prediction so coverage is controlled within each diagnostic class.

• Dual-gate OOD detection combining Mahalanobis distance with a logit-energy criterion
• Mondrian per-class conformal abstention so the model declines rather than guess on out-of-domain cases
• 5-model deep ensemble over RigoBERTa Clinical for the 9-class differential

Manuscript currently under peer review at the Journal of Eukaryotic Microbiology. The full title and abstract will be released when the paper enters the public record.

Manuscript currently under peer review at the Journal of Eukaryotic Microbiology. The full title and abstract will be released when the paper enters the public record.

Matched HeLa+Nf vs Nf-only readouts across LDH (24 h), Caspase-3 (48 h), and JC-1 (72 h) to surface organelle-level dependencies that predict selective amoebicidal effect.

• Plate-level assay design across three timepoints with matched co-culture controls
• Reproducible analysis pipeline with documented per-plate QC

A serum-free, low-cost axenic medium that supports robust Naegleria fowleri growth at a fraction of the per-liter cost of standard formulations, with reproducible passage timing across replicates.

• Three to five times increase in trophozoite yield versus the baseline reference medium
• Cost-aware formulation with documented sourcing for each component
• Reproducible protocol with QC checkpoints and stable passage cadence