New preprint led by @martibartfast. Problem: how to decontaminate SARS-CoV-2 fastq files to remove human reads. Our tool ReadItAndKeep (a C++ wrapper for minimap) removes
human reads in ~1 minute, 10Mb RAM. That's the main result. Thread (details) 1/n
biorxiv.org/content/10.110… To remove host reads, you can broadly either 1. exclude reads that map to the host (human) genome,
or 2. include reads that map to the viral genome 2/n

Last paper, which is a collaboration between the WHO, CRyPTIC and FIND and others, describes the new WHO catalogue of resistance mutations. Preprint here:
papers.ssrn.com/sol3/papers.cf…
1/5 The WHO provides guidance on therapeutic regimens, diagnostic impact of particular mutations and global monitoring of TB. Earlier this year they released their first catalogue of resistance mutations. This preprint describes the methodology 2/5

Next CRyPTIC paper: using machine learning to predict drug resistance/susceptibility, by Alexander Lachapelle and Tim Walker. 1/n Highlights: training an ML model which combined an extreme gradient boosting method based on kmer features, with a catalogue-based prediction where genetic variation across target genes was treated as a signature 2/n

Next CRyPTIC paper: Lindsay Sonnenkalb, @stefan_niemann and colleagues studied bedaquiline (BDQ) and clofazimine (CFZ) resistance, via experimental evolution, studying the structure of Rv0678, and comparison with population data from CRyPTIC. 1/6 Previously, both in vitro and in-patient, a scattering of mutations in Rv0678 had been shown to cause resistance to both BDQ and CFZ. In this study Sonnenkalb evolved Mtb under sub-lethal drug concentrations, to allow a large no. of generations under moderate pressure. 2/6

My group has focussed on the genome analysis for CRyPTIC, and we are very happy to present this study on a new methodology developed by @martibartfast, with an application to the rifampicin resistance :
doi.org/10.1101/2021.0…
A thread (figure to tempt you - see below) 1/18 @martibartfast and I started working on Cryptic back in 2017 or 18 (I forget). Our major concern was how to analyse a huge cohort of up tens of thousands of Mtb and get high recall (needed to detect important alleles) and precision, and jointly genotype the cohort. 2/18

Next CRyPTIC paper:
biorxiv.org/content/10.110…
Josh Carter, @stat_sarah, @philipwfowler modelled MIC w/ multivariable LMM incl. all vars in target genes, controlling for pop struct + id of the pheno lab.Highlight: 449 MIC-elevating muts & 91 muts causing hypersensitivity 1/22 In contrast to the paper of Sarah Earle and @apemandan, which looks genome-wide, hypothesis-free, for associations with MIC, this paper starts with genes known to cause resistance, and looks for effects on MIC. 2/22

Sarah Earle and Daniel Wilson (@apemandan) performed both oligonucleotide and oligopeptide (DNA and amino k-mer) GWAS for 13 drugs, looking for mutations associated with increase or decrease in MIC. Highlights (a thread) 1/n First: this study works off 10k samples, a subset of the full release (because based off an earlier internal release), covering the 4 main Mtb lineages. MIC distributions shown here. Notice small numbers of R for new and repurposed drugs:66 BDQ, 97 CFZ,77 DLM,67 LZD 2/n

Very happy to start this thread-of-threads, on 8 preprints+data-release for @crypticproject, many years of work from >100 collabs collecting,sequencing, phenotyping & analysing >15k M. tuberculosis isolates from 22 countries to understand the genetics of drug resistant TB! 1/n Central to all of our work is the data, a MASSIVE effort to collect and sequence, and then systematically phenotype for drug resistance to 13 drugs using a previously validated (and published) 96-well micro titre plate (reference at end). 2/n

Delighted to see our latest paper now available on biorxiv:
biorxiv.org/content/10.110…
In this study, we set out to solve the problem of SNP calling in bacterial pan-genomes. What problem, you say? .... 1/n What problem, you say? Roughly speaking, given a set of bacterial genomes, there has been no way to detect all the SNPs between them, including accessory genes. We solve this.