A modified decision tree approach to improve the prediction and mutation discovery for drug resistance in Mycobacterium tuberculosis

Background Drug resistant Mycobacterium tuberculosis is complicating the effective treatment and control of tuberculosis disease (TB). With the adoption of whole genome sequencing as a diagnostic tool, machine learning approaches are being employed to predict M. tuberculosis resistance and identify underlying genetic mutations. However, machine learning approaches can overfit and fail to identify causal mutations if they are applied out of the box and not adapted to the disease-specific context. We introduce a machine learning approach that is customized to the TB setting, which extracts a library of genomic variants re-occurring across individual studies to improve genotypic profiling. Results We developed a customized decision tree approach, called Treesist-TB, that performs TB drug resistance prediction by extracting and evaluating genomic variants across multiple studies. The application of Treesist-TB to rifampicin (RIF), isoniazid (INH) and ethambutol (EMB) drugs, for which resistance mutations are known, demonstrated a level of predictive accuracy similar to the widely used TB-Profiler tool (Treesist-TB vs. TB-Profiler tool: RIF 97.5% vs. 97.6%; INH 96.8% vs. 96.5%; EMB 96.8% vs. 95.8%). Application of Treesist-TB to less understood second-line drugs of interest, ethionamide (ETH), cycloserine (CYS) and para-aminosalisylic acid (PAS), led to the identification of new variants (52, 6 and 11, respectively), with a high number absent from the TB-Profiler library (45, 4, and 6, respectively). Thereby, Treesist-TB had improved predictive sensitivity (Treesist-TB vs. TB-Profiler tool: PAS 64.3% vs. 38.8%; CYS 45.3% vs. 30.7%; ETH 72.1% vs. 71.1%). Conclusion Our work reinforces the utility of machine learning for drug resistance prediction, while highlighting the need to customize approaches to the disease-specific context. Through applying a modified decision learning approach (Treesist-TB) across a range of anti-TB drugs, we identified plausible resistance-encoding genomic variants with high predictive ability, whilst potentially overcoming the overfitting challenges that can affect standard machine learning applications.

Evaluation of whole-genome sequence data analysis approaches for short- and long-read sequencing of Mycobacterium tuberculosis

Whole-genome sequencing (WGS) of Mycobacterium tuberculosis (MTB) isolates can be used to get an accurate diagnosis, to guide clinical decision making, to control tuberculosis (TB) and for outbreak investigations. We evaluated the performance of long-read (LR) and/or short-read (SR) sequencing for anti-TB drug-resistance prediction using the TBProfiler and Mykrobe tools, the fraction of genome recovery, assembly accuracies and the robustness of two typing approaches based on core-genome SNP (cgSNP) typing and core-genome multi-locus sequence typing (cgMLST). Most of the discrepancies between phenotypic drug-susceptibility testing (DST) and drug-resistance prediction were observed for the first-line drugs rifampicin, isoniazid, pyrazinamide and ethambutol, mainly with LR sequence data. Resistance prediction to second-line drugs made by both TBProfiler and Mykrobe tools with SR- and LR-sequence data were in complete agreement with phenotypic DST except for one isolate. The SR assemblies were more accurate than the LR assemblies, having significantly (P<0.05) fewer indels and mismatches per 100 kbp. However, the hybrid and LR assemblies had slightly higher genome fractions. For LR assemblies, Canu followed by Racon, and Medaka polishing was the most accurate approach. The cgSNP approach, based on either reads or assemblies, was more robust than the cgMLST approach, especially for LR sequence data. In conclusion, anti-TB drug-resistance prediction, particularly with only LR sequence data, remains challenging, especially for first-line drugs. In addition, SR assemblies appear more accurate than LR ones, and reproducible phylogeny can be achieved using cgSNP approaches.

HIV-1 drug resistance profiling using amino acid sequence space cartography

Motivation: Human immunodeficiency virus (HIV) drug resistance is a global healthcare issue. The emergence of drug resistance demands treatment adaptation. Computational methods predicting the drug resistance profile from genomic data of HIV isolates are advantageous for monitoring drug resistance in patients. Yet, the currently existing computational methods for drug resistance prediction are either not suitable for complex mutational patterns in emerging HIV strains or lack interpretability of prediction results which is of paramount importance in clinical practice. Hence, to overcome these limitations, new approaches for the HIV drug resistance prediction combining high accuracy and interpretability are required. Results: In this work, a new methodology for the analysis of protein sequence data based on the application of generative topographic mapping was developed and applied for HIV drug resistance profiling. It allowed achieving high accuracy of resistance predictions and intuitive interpretation of prediction results. The developed approach was successfully applied for the prediction of HIV re-sistance towards protease, reverse-transcriptase and integrase inhibitors and in-depth analysis of HIV resistance-inducing mutation patterns. Hence, it can serve as an efficient and interpretable tool to suggest optimal treatment regimens. Availability: https://github.com/karinapikalyova/ISIDASeq

Activable Nanoparticle for Tumor Aggressiveness and Drug Resistance Prediction by Glutathione Heterogeneous Imaging

Studies have shown that a higher GSH level is related to more drug-resistant and invasiveness of a tumor. However, it is a great challenge to accurately imaging the GSH level in vivo, for its imaging intensity will interfered by different accumulation of probes in the tumor. Thus, we hypothesized ratiometric photoacoustic imaging that can be used to predict the drug-resistant and invasiveness of tumors by accurate GSH level imaging. In this study, we synthesized MnO2/Indocyanine Green (MnO2/ICG). It can be used as ratiometric photoacoustic (PA) imaging probe, for its absorption at 780 nm (Ab780) can be decreased and 680 nm (Ab680) remain unchanged upon GSH reduction. And our results confirmed that a lower PA absorption ratio (Ab780/Ab680) corresponded to a higher GSH level of the tumor. Besides, the near-infrared fluorescence (FI) imaging was used as an assistant, for it can be quenched upon GSH. Therefore, MnO2/ICG can predict the tumor invasiveness and drug resistance by imaging the GSH level. This study provides reference to predict the prognosis of tumors by imaging the metabolic biomarker of tumors.

Integrated Metabolomics to Reveal the Impacts of Common Antibiotics based on Drug Resistance Prediction of Gut Microbiota

Due to the close interaction between the host and the gut microbiota, the alterations in gut microbiota metabolism may therefore contribute to various diseases. How to use antibiotics more wisely in clinical practice is a promising task in the field of pathophysiology related to gut microbiota. The hope fueling this research is that the alteration of gut microbial communities are paralleled by their capacity on metabolomic from the combined perspective of microbiome and metabolomics. In order to reveal the impacts of antibiotics on microbiota-associated host metabolomic phenotypes, a feasible methodology should be well developed to assess the pervasive effects of antibiotics on the population structure of gut microbial communities. Our attempt starts from predicting specific resistance phenotypes of the individuals in isolation from the rest of the gut microbiota community, according to their resistant genotypes. Once resistance phenotypes of microbiome is determined, we integrated metabolomics with machine learning by applying various analysis algorithms to explore the relationship between the predicted resistance and metabolites, including what the microbial community is after medication, which microbes produce metabolites, and how these metabolites enrich.

Genetic Identification and Drug-Resistance Characterization of Mycobacterium tuberculosis Using a Portable Sequencing Device. A Pilot Study

Clinical management of tuberculosis (TB) in endemic areas is often challenged by a lack of resources including laboratories for Mycobacterium tuberculosis (Mtb) culture. Traditional phenotypic drug susceptibility testing for Mtb is costly and time consuming, while PCR-based methods are limited to selected target loci. We herein utilized a portable, USB-powered, long-read sequencing instrument (MinION), to investigate Mtb genomic DNA from clinical isolates to determine the presence of anti-TB drug-resistance conferring mutations. Data analysis platform EPI2ME and antibiotic-resistance analysis using the real time ARMA workflow, identified Mtb species as well as extensive resistance gene profiles. The approach was highly sensitive, being able to detect almost all described drug resistance conferring mutations based on previous whole genome sequencing analysis. Our findings are supportive of the practical use of this system as a suitable method for the detection of antimicrobial resistance genes, and effective in providing Mtb genomic information. Future improvements in the error rate through statistical analysis, drug resistance prediction algorithms and reference databases would make this a platform suited for the clinical setting. The small size, relatively inexpensive cost of the device, as well as its rapid and simple library preparation protocol and analysis, make it an attractive option for settings with limited laboratory infrastructure.