Typhi Mykrobe: fast and accurate lineage identification and antimicrobial resistance genotyping directly from sequence reads for the typhoid fever agent Salmonella Typhi

Key results

1. 01

   ShownOn Illumina data, read-level k-mer genotyping almost always gave the same GenoTyphi lineage calls as the mapping-based pipeline. Where the two genuinely disagreed, phylogenetic placement and read support favoured the Mykrobe call.

   Same single genotype for 12,807 of 12,834 Typhi genomes (99.79%); 17 more agreed on mixed calls; eight genotype 3.5.3 genomes correctly called by Mykrobe but reported as 3.5.4 by the mapping pipeline; two genuine discordances (ERR5243665, ERR2663783) where Pathogenwatch trees supported Mykrobe; five readsets typed not Typhi. · Validation of GenoTyphi genotype calling; Validation of genotype calls; Tables

2. 02

   ShownAMR genes, resistance mutations and plasmid replicons detected from reads matched Pathogenwatch BLAST calls on assemblies of the same isolates. Drug-level categories and multidrug-resistant (MDR) or extensively drug-resistant (XDR) status also matched.

   n=11,992 isolates; 99.88% of 299,004 marker-genome combinations concordant; per AMR marker 98.56-100%, plasmid replicons 99.74-100%; drug-level agreement >99.7% (2 to 33 discordant genomes per drug); MDR 99.86% (17 discordant), XDR 99.97% (4 discordant). · Validation of AMR genotyping; Tables

3. 03

   ShownAgainst reference-laboratory MIC data, S/I/R predictions from the genotype met US FDA accuracy thresholds for new susceptibility tests for nearly all drugs. Agreement with disk diffusion data from multiple laboratories was lower but still above 95%.

   MIC sets (UKHSA n=852, US CDC n=720): >99% agreement except US CDC low-level ciprofloxacin 97.36% (17 major errors; 99.4% within one doubling dilution); five UKHSA trimethoprim-sulfamethoxazole very major errors (sul, no dfrA). SEAP disk diffusion (n=2,446): 95.3-98.2%. FDA targets: >90% agreement, <3% major, <1.5% very major errors. · Validation of AMR phenotype prediction; Accuracy of AMR phenotype prediction; Discussion

4. 04

   ShownLineage, AMR and plasmid calls from nanopore reads mostly matched Illumina calls for the same isolates. Failures clustered in low-depth data or data from older R9 flowcells.

   98 ONT/Illumina pairs; five ONT sets (R9, mean depth 0.04× to 27.85×) called non-Typhi. Of 92 Typhi genomes: lineage matched 89 (96.7%), with all three misses from R9 runs and missing markers; AMR 98.77% over 736 isolate-drug pairs (ciprofloxacin 91%, chloramphenicol 98.9%); plasmids 99.67% over 1,196 isolate-marker pairs. · Validation on nanopore reads; Matched ONT and Illumina validation data; Discussion

5. 05

   ShownThe tool is fast and memory-light enough for routine local use. It returns a full profile for each genome in under a minute.

   Benchmark of 100 genomes on two HPC systems (1 to 64 threads) and a Mac laptop (1 to 16 threads): <1 minute per genome on a modern computer, down to seconds with multiple threads on a fast CPU, gains up to ~4 threads, typically <100 MB RAM per genome (Supplementary Figure 2). · Benchmarking of run time for Mykrobe; Validation of GenoTyphi genotype calling