TY  - GEN
ID  - heidok30921
TI  - Analysis of high-throughput screening data for precision oncology and antibiotic combination modelling
Y1  - 2021///
AV  - public
CY  - Heidelberg
N2  - The recent success of targeted anticancer therapeutics has propelled cancer genomics to the forefront of clinical oncology. Patients with actionable mutations can now be treated with compounds that target cancer-specific pathways while minimizing damage to healthy tissues. Precision oncology aims to match treatment to a tumor?s mutational composition. Beyond genomic sequencing technologies, drug sensitivity assays play a crucial role in functional profiling of cancers. Tumor sample accessibility makes blood cancers particularly amenable to ex-vivo drug screening. Given the number of functional assays established in leukemias and lymphomas, critics often question the diagnostic value of ex-vivo drug sensitivities.
One limitation of compound screening techniques in liquid cancers is the omission of microenvironment signals secreted by the bone marrow in vivo. To assess the impact of the microenvironment on drug response, I analyzed the high-throughput imaging data obtained from a compound screen conducted in primary leukemias and lymphomas (n = 108). In this study, patient-derived cancer cells were exposed to compound perturbations both alone and in coculture with a bone marrow stromal cell line. In total, 50 compounds were probed at 3 different concentrations across 2 culture conditions. I developed an automated analysis workflow, which was applied to > 700,000 confocal microscopy images. To extract multivariate phenotypes, I implemented a Python package (bioimg) for single-cell morphological profiling and performed the statistical analysis of compound effects in mono- and coculture.
One of the key findings of the leukemia-stroma coculture study was that about 50% of the probed compounds were less effective in coculture compared to monoculture. Stratifying by compound class, I found that the efficacy of chemotherapeutics, BET and proteasome inhibitors was diminished by stroma-mediated protection. JAK inhibitors were the only compounds in the screen that reduced stromal protection. However, pro-survival effects of stroma were not uniform and stroma-induced morphology changes observed in cancer cells varied among samples. To understand this variability, I explored drug-gene associations in the presence of microenvironment signals and found that IGHVunmutated and trisomy-12-positive samples gained stronger stromal protection when treated with BCR inhibitors.
In addition to precision medicine, antibiotic resistance research makes common use of highthroughput screening techniques. Misuse of antibiotics is driving the evolution and expansion of antibiotic-resistant pathogens, posing a significant risk to global public health. The current drug development efforts to combat this urgent threat are inadequate outside of academia. To address this gap, researchers have developed drug combination profiling systems to identify synergistic and antagonistic drug pairs. Species and strain specificity of synergies and antagonisms necessitates high-throughput investigations in multiple bacterial strains. To tackle these challenges, I analyzed the largest antibiotic combination screen in Gram-positive bacteria that exists to date. In this study, about 2000 drug pairs were probed in B. subtilis, S. pneumoniae, and two S. aureus strains. I developed a computational analysis pipeline that processed high-throughput bacterial growth data. My analysis revealed the landscape of drug interactions in Gram-positive species. I observed high within-class synergistic rates, especially for cell wall targeting compounds and inhibitors of protein and DNA synthesis. Compared with drug interactions observed in Gram-negatives, both abundance and interspecies conservation rates of synergies and antagonisms were lower in the Gram-positive organisms.
Currently, high-throughput screening remains the only feasible method of uncovering drug-drug interactions on a large scale, as only a minority of synergy and antagonism mechanisms have been fully elucidated. To facilitate the rational design of combinatorial therapies, I analyzed drug-drug and druggene interaction data in E. coli and S. typhimurium with the goal of identifying genetic determinants of drug-drug interactions. Consistent with the previous findings in E. coli, my analysis revealed that ATP and lipopolysaccharide biosynthesis were among the most important biological processes for drug-drug interactions. In E. coli I found that ATP synthesis was particularly important for synergistic interactions among cell wall targeting compounds. Furthermore, I could show that it is possible to predict novel drug-drug interactions using chemogenomic data in E. coli and S. typhimurium. Finally, using a trained machine learning model I was able to identify genes predictive of drug-drug interactions within specific drug classes and calculate feature importance for individual predictions.
UR  - https://archiv.ub.uni-heidelberg.de/volltextserver/30921/
A1  - Kim, Vladislav
ER  -