Why modelling patient heterogeneity is crucial in drug development
Prof Shai Shen-Orr at CytoReason explores the need for patient heterogeneity in drug development, and suggests why this is so central to successful drug development
development is heavily dependent on our understanding of disease mechanisms, and patient heterogeneity is crucial for achieving improved modelling of the complex disease landscape – often reflected in diverse disease clinical phenotypes and severity levels. Prevalent approaches such as the use of cell lines, organoids and animal models, have limitations in accurately representing the complexity of human diseases and the diversity among patients. However, advancements in artificial intelligence (AI) have provided new opportunities to overcome some of these challenges. By accounting for patient-specific characteristics, computational disease models can provide a tailored and precise description of disease mechanisms, drug responses, adverse effects and potential therapeutic targets, enhancing personalised medicine and effective drug discovery strategies.
Patient heterogeneity modelling supports targeted drug development by patient stratification and identification of patients’ sub-populations that are most likely to respond to a particular drug. This enables drug development efforts to focus on those patients and to enhance clinical trial design, leading to a more efficient development process and an increased likelihood of success. Ultimately, this approach promotes better patient outcomes and reduced costs related to trial and error procedures, which are highly time-consuming and expensive. Through stratifying patients based on their disease landscape and focusing on more homogenous cohorts, a greater number of genes associated with the mechanism of action (MOA) emerge. This also holds significant relevance in identifying drug resistance mechanisms.
Drug repurposing is an additional factor enhanced by identifying existing drugs that are effective in treating patient subgroups, originally not targeted by those drugs. The variation among patients can affect drug pharmacodynamics that is related to drug activity, and pharmacokinetics that includes aspects of drug absorption, metabolism, distribution and excretion. All of these can contribute to different outcomes in drug therapy. A broad understanding and precise mapping of drug MOAs, and specific characteristics across indications of patient sub-populations, may enable better matching of existing drugs for new indications, and thereby accelerate drug development and reduce costs. Beyond this, patient heterogeneity is also important for drug toxicity assessment, through the identification of sub-populations expressing distinct features causing high predisposition to increased risk of drug toxicity.
While patient heterogeneity aims to identify variations observed among real patients, another complementary approach uses virtual patients and digital twins to focus on simulating a dynamic virtual representation of an individual to better understand and predict health outcomes at the patient level. This is particularly useful in making informed treatment decisions, monitoring and optimising interventions for a specific patient. While both approaches are valuable, the former prioritises disease-oriented research, whereas the latter emphasises patient insights.
There are multiple approaches to model patient heterogeneity, which places molecular omics data, together with machine learning (ML) tools including supervised, unsupervised and reinforcement learning, at the forefront of pharma research. Expanding the ability to generate patient-specific data advances the use in ML approaches for drug development and treatment decision-making. Extensive and high-resolution types of patient data are measured for these purposes including genomics, transcriptomics, metabolomics, epigenetics, microbiome, spatial and temporal data and electronic health records (EHR).
The computational modelling of patient heterogeneity in drug development offers a powerful approach to enhance our understanding of disease complexity, optimise treatment strategies and accelerate the discovery of effective therapies
While genetic heterogeneity is widely recognised, patient heterogeneity extends beyond genetics
and is influenced by various additional factors, including environmental exposures, lifestyle choices, comorbidities and demographic factors. Disease development and progression encompass a continuum of changes occurring in numerous parameters, covering both intra- and inter-individual variations in cellular composition and regulatory programmes over time and space. To capture and represent this multifaceted diversity, trajectory inference methods have been used, which enable the pinpointing of factors that drive state transitions and putative causal drivers of the onset and progression of the disease. Dissection of these system-wide effects substantially helps designing new drug targets and treatment manipulations to modulate the identified disease drivers.
Systems biology and network-based approaches are powerful tools in the study of disease biology. While many network analyses rely on population-level averages, recent advancements have enabled the creation of sample-specific networks that can capture the complex molecular interactions present at the individual level. These networks allow identifying key signaling pathways that are dysregulated in a particular patient, leading to the discovery of potential therapeutic targets that are specific to a patient sub-population. Drug-response networks can be constructed using patient-specific data on drug treatments. These networks can help identify drug responses, biomarkers and drug resistance mechanisms, empowering optimisation of treatment strategies accordingly.
Bayesian modelling, which involves developing probabilistic models to predict disease and drug responses, is used to model patient heterogeneity by incorporating prior knowledge and uncertainty into the analysis. Bayesian modelling can help identify patient subgroups and their associated properties, and support decision-making related to treatment selection. For large sized data sets, as well as complex unstructured data such as documents, images and text, including medical imaging and electronic health records, recently deep learning algorithms have been used to model disease heterogeneity, in order to predict the risk of disease development or patient-specific disease outcomes.
Taken together, the computational modelling of patient heterogeneity in drug development offers a powerful approach to enhance our understanding of disease complexity, optimise treatment strategies and accelerate the discovery of effective therapies. This comprehensive approach holds great promise for advancing personalised medicine and transforming the field of drug development.
Systems biologist and data scientist
is the co-founder and chief scientist of
and a professor in the faculty of medicine at the Technion, where he directs the laboratory of systems immunology and precision medicine. In his research, he develops new analytical methodologies for grappling with the intricate complexities of the immune system; his research has been cited numerous times and has been featured in systems biology textbooks for students. Shen-Orr received a BSc from the Technion in Information Systems, an MSc in Bioinformatics at the Weizmann Institute of
Science and a PhD in Biochemistry from Harvard University. He performed his postdoctoral studies at