Network pharmacology is a new field of science focused on targeting multiple steps in a regulatory signaling network. The goals of this field include facilitating the design of drugs with specific multi-target profiles and exploiting the existing polypharmacology of many currently used medicines. Given that kinases represent one of the largest target families in drug development and critical components of all signaling networks, we are developing computational tools for evaluating potential clinical applications of kinase inhibitors. Through these efforts, we aim to enhance our understanding of the basic kinase biology as well as advance pharmacological exploitation of these key cellular regulators. Our lab has established a series of machine learning approaches that use large scale drug-target profiling efforts, machine learning approaches, and broadly-selective chemical tool compounds to pinpoint specific nodes (kinases and associated networks) underlying a given phenotype such as the growth of cancer cells or release of cytokines. Using a combination of these approaches, we are able to 1) identify specific signaling nodes that are important for a given phenotype; 2) predict response to FDA-approved or clinical grade drugs as single agents and rank order drug combinations in silico, and 3) build kinase-centered network-level models to identify critical nodes and identify optimal drug combinations for any given phenotype. In the past few years, we have applied a combination of these systems-based approaches to broad areas of biology, ranging from the studies of malaria, COVID-19, and cancer . These studies identified new molecular regulators and potential therapeutics, highlighting the potential of these computational tools for unbiased biological discovery. Overall, our lab is a multidisciplinary and collaborative research environment that aims to be at the forefront of the new field of network pharmacology and deliver on its potential to dramatically accelerate the discovery of new biological insights and their translation into curative therapies. Read More
Kinase Inhibitor Regularization (KiR): By applying machine-learning approaches to pharmacological response data, we are identifying kinases that participate in specific cellular processes, such as EMT, cell migration, or cell differentiation. Most kinase inhibitors have broad specificity; they inhibit multiple types of kinases. However, the exact set of kinases blocked by each inhibitor is different. This property of broad, but nonidentical, specificity is the basis for “Kinase Inhibitor Regularization” (KiR). We used KiR to identify kinase targets of inhibitors using data from the effect of the inhibitors on the behavior of cultured cells, a phenotypic screen. In one application of KiR, we screened a set of kinase inhibitors, designed using computational methods for optimal specificity, in a panel of cancer cell lines for those inhibitors that reduced or enhanced cell migration. We identified cancer cell line-specific kinases that regulate cell migration. My lab continues to use KiR to explore kinases in phenotypes associated with cell growth, differentiation, and migration. This approach can also be used to predict cell type-specific responses to kinase inhibitors.
The ever-increasing size and scale of biological information have created a need for systems-level tools that synthesize large quantities of dispersed and distinct data and inform decision- and hypothesis-making processes. To address this need, we have created KiRNet, a kinase-centered method designed to integrate results of functional screens with protein-protein interaction data, and additional molecular data. KiRNet produces functional, network-level models that are optimized and refined to identify small, differentially regulated subnetworks, even in the absence of large-scale datasets. As a proof of concept, we applied KiRNet to liver cancer cells overexpressing FZD2, a gene known to drive the epithelial-mesenchymal transition and cancer metastasis and identified a subnetwork of 166 proteins that regulate this cell state. We demonstrate that KiRNet can formulate high-value predictions for future testing and thus accelerate basic and translational discoveries.
KiRNet: Kinase-centered network propagation of pharmacological screen results, Cell Reports Methods, 2021
Access Publicly-accessible repository for KiRNet Script here
Because of the broad specific of most kinase inhibitors, selecting an appropriate inhibitor for a given kinase is challenging. We developed an online portal, KInhibition at which users can search publicly available datasets to find selective inhibitors for a kinase or group of kinases. Compounds are sorted according to a KInhibition Selectivity Score, which is calculated based on the activity of the compound against the selected kinase(s) versus the activity against all other kinases for which that compound has been profiled. KInhibition represents a powerful platform through which researchers with diverse interests can easily interrogate large datasets to help guide their selection of kinase inhibitors. My lab maintains this online resource for the community and will make improvements as more datasets become available and existing datasets are updated. Access KInhibition here
"KInhibition: A Kinase Inhibitor Selection Portal." iScience, 2018
"Making the right move with a KISS." Science Spotlight, 2018
Application of deep learning algorithms has the potential to reduce the overall costs of pre-clinical drug development and accelerate discovery of high-value lead compounds. Our labs has recently developed KiDNN (Kinase Inhibitor prediction using Deep Neural Networks) to predict phenotypic effects of kinase inhibitors. Unlike previous studies that used linear networks to model kinase signaling, we now introduce a non-linear, multilayer feed-forward network that more closely mimic complex and dynamic nature of kinase-driven signaling pathways. We used KiDNN to predict the effect of ~200 kinase inhibitors on migration of cultured breast and liver cancer cells as a model phenotype. We compared prediction accuracy of KiDNN to other prediction tools based on linear models and determined through experimental testing that KiDNN outperformed the linear models. Further, we show that DNN hyperparameters learned from one set of data (breast cancer cells) can be used to generate KiDNN that predicts responses in multiple, unrelated cancer models with minimal training data. Overall, non-linear, DNN-based models provide a powerful approach to in silico screen hundreds of kinase inhibitors. Our work further supports the potential of deep learning algorithms to accelerate the discovery of lead compounds for subsequent development of drug candidates.