Antivirals against the PEDV spike protein, from screen to bench
Master's thesis, NPUST, 2024 to 2026
In silico screening and wet laboratory validation
Question. Can structure based virtual screening identify compounds that inhibit Porcine
Epidemic Diarrhea Virus, and do those compounds hold up when tested in cells?
Method. High throughput docking of compound libraries against the spike protein,
molecular dynamics on the leading poses, then cytotoxicity and virus inhibition assays in Vero
cells.
Screen to assayCandidates carried through the
full sequence to measured viral replication. Presented at ISSI-SA 2026 as first author.
Two annotators, one gene caller, half the product names different
bakta_prokka_disagreement_prediction
Python, random forest
Question. Bakta and Prokka both wrap Prodigal, so they agree on where genes begin and
end. Where does the annotation diverge, and is that divergence systematic enough to predict?
Method. Comparison of 87,859 coding regions across 25 complete bacterial genomes, with a
classifier trained on sequence composition and genomic context.
51.7 per centof identically bounded coding
regions receive different product names, and the disagreement is predictable rather than random.
Repository
Ranking colorectal cancer risk beyond tumour stage
cancer_survival_ml_pipeline
Python, gradient boosting against a Cox baseline
Question. Does gradient boosting order patients by risk of death more accurately than
tumour stage alone, and by a margin large enough to be clinically interesting?
Method. 586 patients with 185 events, with the proportional hazards model retained as the
reference rather than discarded once the machine learning result proved higher.
0.729 against 0.674concordance index,
gradient boosting compared with stage. Both figures are reported together.
Repository
Does TMAO carry information after a coronary intervention?
tmao_cardiovascular_risk_prediction
Python, pre-registered analysis
Question. Does trimethylamine N-oxide, a metabolite of gut microbial origin, predict
recurrent angina beyond what choline, betaine and carnitine already explain?
Method. Reanalysis of 750 patients following percutaneous coronary intervention. The
outcome definition, covariates, calibration procedure and batch effect handling were written,
dated and published before the data were examined.
750 patientsanalysed against a
pre-registration that cannot be revised once the result is known.
Repository
The within host variome of a single influenza infection
nf-core-viralrecon-influenza
Nextflow, Bowtie2, iVar
Question. How much quasispecies diversity does one infection carry, and where does that
diversity sit across a segmented genome?
Method. An eight segment supercontig reference built from A/California/07/2009, with
reference guided alignment and minor variant calling at 906.97 fold mean depth, executed within
a 3.7 gigabyte memory ceiling.
48 variants, Ts/Tv 7.8The transition bias
expected of an RNA dependent RNA polymerase, measured rather than assumed.
Repository
Penicillin resistance across 604 pneumococcal isolates
pyseer_bgwas_analysis
Shell, k-mer bacterial GWAS
Question. Which genetic determinants drive penicillin resistance in
Streptococcus pneumoniae once clonal background has been properly accounted for?
Method. k-mer based genome wide association with pyseer across 604 clinical isolates,
using an explicit lineage model to control for population structure.
604 isolatesDeterminants that survive
correction for population structure, rather than determinants that merely track it.
Repository
Qasim Hussain