Drug target prediction through deep learning functional representation of gene signatures.

Many machine learning applications in bioinformatics currently rely on matching gene identities when analyzing input gene signatures and fail to take advantage of preexisting knowledge about gene functions. To further enable comparative analysis of OMICS datasets, including target deconvolution and mechanism of action studies, we develop an approach that represents gene signatures projected onto their biological functions, instead of their identities, similar to how the word2vec technique works in natural language processing. We develop the Functional Representation of Gene Signatures (FRoGS) approach by training a deep learning model and demonstrate that its application to the Broad Institute's L1000 datasets results in more effective compound-target predictions than models based on gene identities alone. By integrating additional pharmacological activity data sources, FRoGS significantly increases the number of high-quality compound-target predictions relative to existing approaches, many of which are supported by in silico and/or experimental evidence. These results underscore the general utility of FRoGS in machine learning-based bioinformatics applications. Prediction networks pre-equipped with the knowledge of gene functions may help uncover new relationships among gene signatures acquired by large-scale OMICs studies on compounds, cell types, disease models, and patient cohorts.

Structure-activity relationship of dihydropyridines for rhabdomyosarcoma.

Childhood muscle-related cancer rhabdomyosarcoma is a rare disease with a 50-year unmet clinical need for the patients presented with advanced disease. The rarity of ∼350 cases per year in North America generally diminishes the viability of large-scale, pharmaceutical industry driven drug development efforts for rhabdomyosarcoma. In this study, we performed a large-scale screen of 640,000 compounds to identify the dihydropyridine (DHP) class of anti-hypertensives as a priority compound hit. A structure-activity relationship was uncovered with increasing cell growth inhibition as side chain length increases at the ortho and para positions of the parent DHP molecule. Growth inhibition was consistent across n = 21 rhabdomyosarcoma cell line models. Anti-tumor activity in vitro was paralleled by studies in vivo. The unexpected finding was that the action of DHPs appears to be other than on the DHP receptor (i.e., L-type voltage-gated calcium channel). These findings provide the basis of a medicinal chemistry program to develop dihydropyridine derivatives that retain anti-rhabdomyosarcoma activity without anti-hypertensive effects.

Discovery of small molecules that target a tertiary-structured RNA.

There is growing interest in therapeutic intervention that targets disease-relevant RNAs using small molecules. While there have been some successes in RNA-targeted small-molecule discovery, a deeper understanding of structure-activity relationships in pursuing these targets has remained elusive. One of the best-studied tertiary-structured RNAs is the theophylline aptamer, which binds theophylline with high affinity and selectivity. Although not a drug target, this aptamer has had many applications, especially pertaining to genetic control circuits. Heretofore, no compound has been shown to bind the theophylline aptamer with greater affinity than theophylline itself. However, by carrying out a high-throughput screen of low-molecular-weight compounds, several unique hits were identified that are chemically distinct from theophylline and bind with up to 340-fold greater affinity. Multiple atomic-resolution X-ray crystal structures were determined to investigate the binding mode of theophylline and four of the best hits. These structures reveal both the rigidity of the theophylline aptamer binding pocket and the opportunity for other ligands to bind more tightly in this pocket by forming additional hydrogen-bonding interactions. These results give encouragement that the same approaches to drug discovery that have been applied so successfully to proteins can also be applied to RNAs.

Evolution of Novartis' Small Molecule Screening Deck Design.

This article summarizes the evolution of the screening deck at the Novartis Institutes for BioMedical Research (NIBR). Historically, the screening deck was an assembly of all available compounds. In 2015, we designed a first deck to facilitate access to diverse subsets with optimized properties. We allocated the compounds as plated subsets on a 2D grid with property based ranking in one dimension and increasing structural redundancy in the other. The learnings from the 2015 screening deck were applied to the design of a next generation in 2019. We found that using traditional leadlikeness criteria (mainly MW, clogP) reduces the hit rates of attractive chemical starting points in subset screening. Consequently, the 2019 deck relies on solubility and permeability to select preferred compounds. The 2019 design also uses NIBR's experimental assay data and inferred biological activity profiles in addition to structural diversity to define redundancy across the compound sets.

A Fully Automated High-Throughput Flow Cytometry Screening System Enabling Phenotypic Drug Discovery.

The goal of high-throughput screening is to enable screening of compound libraries in an automated manner to identify quality starting points for optimization. This often involves screening a large diversity of compounds in an assay that preserves a connection to the disease pathology. Phenotypic screening is a powerful tool for drug identification, in that assays can be run without prior understanding of the target and with primary cells that closely mimic the therapeutic setting. Advanced automation and high-content imaging have enabled many complex assays, but these are still relatively slow and low throughput. To address this limitation, we have developed an automated workflow that is dedicated to processing complex phenotypic assays for flow cytometry. The system can achieve a throughput of 50,000 wells per day, resulting in a fully automated platform that enables robust phenotypic drug discovery. Over the past 5 years, this screening system has been used for a variety of drug discovery programs, across many disease areas, with many molecules advancing quickly into preclinical development and into the clinic. This report will highlight a diversity of approaches that automated flow cytometry has enabled for phenotypic drug discovery.

Cytogenetic and genetic pathways in therapy-related acute myeloid leukemia.

Therapy-related myelodysplastic syndrome and acute myeloid leukemia (t-MDS/t-AML) are late complications of cytotoxic therapy used in the treatment of malignant diseases. The most common subtype of t-AML ( approximately 75% of cases) develops after exposure to alkylating agents, and is characterized by loss or deletion of chromosome 5 and/or 7 [-5/del(5q), -7/del(7q)], and a poor outcome (median survival 8 months). In the University of Chicago's series of 386 patients with t-MDS/t-AML, 79 (20%) patients had abnormalities of chromosome 5, 95 (25%) patients had abnormalities of chromosome 7, and 85 (22%) patients had abnormalities of both chromosomes 5 and 7. t-MDS/t-AML with a -5/del(5q) is associated with a complex karyotype, characterized by trisomy 8, as well as loss of 12p, 13q, 16q22, 17p (TP53 locus), chromosome 18, and 20q. In addition, this subtype of t-AML is characterized by a unique expression profile (higher expression of genes) involved in cell cycle control (CCNA2, CCNE2, CDC2), checkpoints (BUB1), or growth (MYC), loss of expression of IRF8, and overexpression of FHL2. Haploinsufficiency of the RPS14, EGR1, APC, NPM1, and CTNNA1 genes on 5q has been implicated in the pathogenesis of MDS/AML. In previous studies, we determined that Egr1 acts by haploinsufficiency and cooperates with mutations induced by alkylating agents to induce myeloid leukemias in the mouse. To identify mutations that cooperate with Egr1 haploinsufficiency, we used retroviral insertional mutagenesis. To date, we have identified two common integration sites involving genes encoding transcription factors that play a critical role in hematopoiesis (Evi1 and Gfi1b loci). Of note is that the EVI1 transcription factor gene is deregulated in human AMLs, particularly those with -7, and abnormalities of 3q. Identifying the genetic pathways leading to t-AML will provide new insights into the underlying biology of this disease, and may facilitate the identification of new therapeutic targets.

Haploinsufficiency of EGR1, a candidate gene in the del(5q), leads to the development of myeloid disorders.

Loss of a whole chromosome 5 or a deletion of the long arm, del(5q), is a recurring abnormality in myelodysplastic syndromes (MDSs) and acute myeloid leukemia (AML). To identify a leukemia-related gene on chromosome 5, we previously delineated a 970-kb segment of 5q31 that is deleted in all patients examined, and prepared a transcript map of this region. EGR1 is a candidate tumor suppressor gene within the commonly deleted segment of 5q, and encodes a zinc finger transcription factor. To test the hypothesis that loss of function of Egr1 is an initiating event in the pathogenesis of AML/MDS, Egr1-deficient mice were treated with a potent DNA alkylating agent, N-ethyl-nitrosourea (ENU), to induce secondary cooperating mutations. Egr1(+/-) and Egr1(-/-) mice treated with ENU developed immature T-cell lymphomas (CD4(+), CD8(+)) or a myeloproliferative disorder (MPD) at increased rates and with shorter latencies than that of wild-type littermates. The MPD was characterized by an elevated white blood cell count, anemia, and thrombocytopenia with ineffective erythropoiesis. Biallelic mutations of Egr1 were not observed in MPDs in Egr1(+/-) mice. Our data suggest that haploinsufficiency for Egr1 plays a role in murine leukemogenesis, and in the development of AML/MDS characterized by abnormalities of chromosome 5.

The gene mutated in variant late-infantile neuronal ceroid lipofuscinosis (CLN6) and in nclf mutant mice encodes a novel predicted transmembrane protein.

The neuronal ceroid lipofuscinoses (NCLs) are a group of autosomal recessive neurodegenerative diseases characterized by the accumulation of autofluorescent lipopigment in various tissues and by progressive cell death in the brain and retina. The gene for variant late-infantile NCL (vLINCL), CLN6, was previously mapped to chromosome 15q21-23 and is predicted to be orthologous to the genes underlying NCL in nclf mice and in South Hampshire and Merino sheep. The gene underlying this disease has been identified with six different mutations found in affected patients and with a 1-bp insertion in the orthologous Cln6 gene in the nclf mouse. CLN6 encodes a novel 311-amino acid protein with seven predicted transmembrane domains, is conserved across vertebrates and has no homologies with proteins of known function. One vLINCL mutation, affecting a conserved amino acid residue within the predicted third hydrophilic loop of the protein, has been identified, suggesting that this domain may play an important functional role.