LSP Research Projects

Kinase inhibitors are an intensively studied class of therapeutics with many remaining unknowns – LSP investigators use them as a laboratory in which to develop new approaches to target identification and study factors determining resistance and sensitivity to existing drugs.

Inhibitors of the roughly 540 human protein and lipid kinases are the most intensively developed and studied class of therapeutic drugs. Kinase inhibitors such as imatinib for treatment of chronic myeloid leukemia (CML), vemurafenib for treatment of BRAF-mutant melanoma, and erlotinib for treatment of EGFR-mutant lung cancer were the first breakthrough  “targeted” drugs invented for genetically defined cancers. Kinase inhibitors are also important in the treatment of inflammatory diseases. A primary limitation in the use of these drugs is the frequent development of resistance, which involves both a short-term adaptive resistance and genetically determined long-term resistance. These resistance mechanisms involve many of the homeostatic processes that allow cells to survive and proliferate in the presence of internal and external stress and are therefore of great interest in their own right. By studying precisely how drug resistance arises in cell lines, animal models, and humans we hope to prevent its emergence using newly developed drugs or combination therapies. This project is ideal for cancer biologists and oncologists interested in fundamentally improving our approach to small molecule therapy and increasing the durability of therapeutic responses. Clinical trials are underway making use of some of the insights arising from this project.

LSP investigators are contributing to and benefiting from the rapid advances in deep learning that are revolutionizing data analysis and protein structure prediction and that promise to fundamentally advance our ability to invent new drugs

The contemporary approach to invention of targeted therapies commonly attempts to block the action of specific disease genes through the development of highly selective inhibitors. In the case of multi-gene families such as kinases it is increasingly evident that approved drugs commonly inhibit multiple proteins and that this poly-pharmacology is important to their mechanism of action. To better understand the spectrum of proteins that drugs bind we are investigating drugs targeting multi-enzyme families (kinases, deubiquitinases, and bromodomain inhibitors) using cell-based assays, mass spectrometry, and deep learning (a type of machine learning and artificial intelligence; (AI). Structure prediction and docking using deep learning plays an increasingly important role and members of our lab are active in building on recent successes in this area (AlphaFold for example). This project is actively recruiting individuals interested in ML/AI, protein structure prediction, and chemical biology.

The LSP has developed new highly multiplexed imaging methods and computational tools to enable deep molecular interrogation of tumor-immune interactions and develop next-generation diagnostics for precision cancer care.

The development of a precise, patient-centered, approach to medicine is critically dependent on being able to more precisely diagnose disease and determine which patients are most likely to benefit from a specific therapy. In oncology, disease diagnosis and  management are dominated by information acquired by pathologists using histology, the examination of tissue and tumor sections in the light microscope. Despite a growing body of knowledge and the increasing impact of genetic information, the practice of pathology remains largely pre-digital. The LSP is at the forefront of developing a new generation of highly-multiplexed tissue imaging methods that can acquire precise single cell molecular information in histological specimens and fuse this with other types of genetic and single-cell data. This approach is fundamentally changing our understanding of tumor immune-interaction and the ways in which therapeutic interventions can be tailored to specific tumor types. The development of AI algorithms and large public databases (as part of the NCI Cancer Moonshot Program) is a key part of this program and we are optimistic that we will have novel methods in clinical use within two to three years. The LSP created the Harvard Tissue Atlas Project in this area as well as the Ludwig Tumor Atlas. Our tissue imaging projects are actively recruiting staff, fellows, and students interested in cancer biology, immunotherapy, microscopy, optical physics, machine learning/AI, and cloud computing.

LSP investigators working with the Massachusetts Alzheimer's Disease Research Center at the MGH are developing new approaches to treat Alzheimer's Disease based on the hypothesis that the disease has multiple distinct etiologies some involving degeneration-associated chronic inflammation.

As populations age, Alzheimer's Disease (AD) poses a rapidly increasing burden for healthcare systems and the disease is a tragedy for the individuals involved and their families.  The discovery of new drugs for neurodegenerative diseases is challenging because the etiology of these diseases is poorly understood, the diseases frequently progress relatively slowly, and end-stage disease – when symptoms are most obvious -  is likely to remain difficult to treat.  For these reasons we are pursuing multiple approaches to repurposing drugs that are approved by the Food and Drug Administration (FDA) for others indications, commonly an inflammatory disease. We are investigating repurposing opportunities by mining large-scale clinical databases (in collaboration with colleagues in the UK) and by developing pre-clinical models of neurodegeneration.  The machine learning algorithm DRIAD (Drug Repurposing In AD), for example, uses profiling data from human neural cell cultures treated with existing compounds to generate ranked lists of possible repurposing candidates. Some of these candidates have advanced to early-stage clinical trials. However, our long-term goal is not to stick with repurposed drugs but instead to use them to better understood disease mechanisms and guide the development of more effective second generation molecules tailored to specific types of Alzheimer's Disease.

Through a close collaboration with Tufts Medical School, LSP scientists are using computational approaches, animal models, and deep profiling of TB granulomas to understand how mechanisms of combination therapy and improve treatment approaches.

Tuberculosis remains a scourge in many parts of the developing world and is also common in disadvantaged and immigrant communities in the US. Treating the disease remains challenging, with too few drugs available, patient-to-patient variability in response, and emergence of drug resistance. LSP scientists are studying TB granulomas  - sites of inflammation - that represent sites of active or residual disease in which the host immune system is attempting to eradicate infection. Granulomas evolve over time in an irregular fashion even within a single patient and can become the source of resurgent disease. Developing better treatments for TB faces many of the same hurdles as developing drugs for cancer and many TB granulomas are discovered in the process of screening for lung cancer. LSP investigators are using a wide range of tools to study granulomas in humans and animal models. In collaboration with the Bill and Melinda Gates Foundation the LSP is also developing an on-line digital resource for the TB community that focuses initially on fusing data from advanced tissue imaging, single-cell sequencing and radiological scans.

LSP investigators are using screens in iPSC-derived neurons, cheminformatics, and sophisticated animal models to develop novel non-opioid treatments for pain with greater long-term efficacy and lower abuse potential than current options.

Existing therapies for pain are often ineffective or liable to abuse, contributing to our current addiction and overdose crisis. However, development of non-opioid analgesics is one of the most challenging problems in contemporary drug discovery. An ambitious multi-institution project funded by DARPA (the Panacea Research Program) at HMS aims to develop novel small molecules and biologics for treatment of pain and inflammation while also developing more effective computation-enabled methods for drug discovery platforms. Our approach is target agnostic with screening centered on cell-based phenotypes that are indicative of nociceptor-specific silencing of channels and other proteins involved in drug sensation.  The project uses a diversity of cutting-edge pharmacology tools, including iPSC-derived neuronal drug screening, mass spectrometry proteomics, protein structure prediction and docking, computation assisted synthesis planning, and advanced AI-assisted behavioral assays. Related efforts are applying similar approaches to both chemotherapy-induced and diabetic peripheral neuropathy. This program is actively recruited medicinal chemists, chemical biologists and neurobiologists.

The LSP is developing new computing platforms and knowledge assembly systems for dramatically improving how we extract causal and mechanistic information from the published literature, most of which is in a text-based form that is not readily computable.

Approaches to studying biological regulation, disease mechanisms, and drug targets that involve systems pharmacology approaches require computable knowledge about biological networks. Many databases have been developed to curate information on gene functions and protein-protein interactions but much of the information in the literature is not yet recorded in databases in a useful way. We are therefore developing novel approaches to knowledge assembly that combine natural language processing (NLP) with novel domain-specific programming languages such as PySB that can capture complex biological concepts in computable form. Much of this work focuses on advancing the Integrated Network and Dynamical Reasoning Assembler (INDRA), an automated model assembly software system that uses NLP to rapidly scan the published literature and also connects to existing databases to collect knowledge from a wide range of existing database. INDRA then assembles this knowledge into a self-consistent form, enabling the generation of causal graphs and dynamical computational models of disease mechanisms and drug action.

The LSP is engaged in a wide-ranging project to understand the causes of success and failure in cancer clinical trials and developing improved computation methods for early-phase (signal finding) trials that are predictive of success in future trials.

Clinical trials are the most expensive and important stage of drug development and yet the determinants of success and failure remain poorly understood. In part this is because it is so hard to access data on the outcomes of existing trials. We are therefore creating a database of survival curves and individual participant data (IPD) that can be used to test specific hypotheses about drug mechanism of action and improve how early phase trials are scored. We have already determined that mechanisms of combination cancer therapy rarely involve drug interaction (synergy)  but are instead based on the mechanism of independent drug action in which each patient benefits from the single agent in the combination with the greatest activity. This approach is already being used by leading pharmaceutical companies to prioritize future immunooncology trials. In related work we have developed methods to increase the statistical power associated with small clinical trials, including the basket trials increasingly used to identify responsive patient populations. Our data suggests that current therapeutic approaches would dramatically benefit from increasing precision based in part on new diagnostics involving next-generation digital histology.