Led by founder and CEO Dr. Jo Varshney, VSL’s team is committed to reducing unnecessary and erroneous (less than 5 percent accurate) animal testing in research and development. Today, at VSL, Dr. Varshney has a team of world-class scientists, machine learning engineers and in silico biosimulation experts who are bringing clarity and confidence to drug development through AI-driven mechanistic modeling and simulation rather than the classic “trial and error” methodology.
VSL’s core technology is a proprietary platform, BIOiSIM, that integrates AI/ML approaches with mechanistic models and provides scalable, translatable, and predictive models to achieve accurate preclinical and clinical outcomes.
Following is the conversation that CIO Applications had with Dr. Varshney to understand how VSL is setting a new standard for drug development.
What are the challenges existing in the drug development space? How is VeriSIM, by leveraging simulation, effectively addressing these issues?
Prior to VSL’s disruption in the pharma space, there was a fundamental disconnect in how medicine gets to treating the patients who need it. Preclinical stages relied heavily on animal testing to understand how the safety and efficacy of a new drug compound will react within the human body.
This process may not be as effective for a few reasons. First, animal physiological outcomes translate poorly in humans during clinical trials, leading to over 92 percent drug failure rate in human trials. Also, identifying a true promising lead may lead to several rounds of animal testing that drives significant time and costs during the drug development process.
To address them, computational resources can build digital models of animal and human responses to drug(s) in parallel to experimental workflows to create scalability, efficiency, and accuracy in the R&D pipeline. As promising as they may seem, such approaches tend to be higher in false positives and negatives rates. These models are also unable to translate accurately from one species to another.
On the other hand, we use world-class expertise in machine learning, mechanistic modeling, and engineering to produce models that serve as ‘digital’ representations of animals, including humans. We integrate the existing knowledge inclusive of physiology, biology, multi-omics, etc. with mathematical equations and several AI/ML models to make them scalable and translatable across several species.
Could you walk us through the BIOiSIM platform and shed light on its features, and benefits involved.
The BIOiSIM platform integrates deep learning/machine learning (DL/ML) with robust in vivo physiological models (across seven species, including humans), enabling highly scalable and faster model development, animal testing reduction (>50 percent), and more accurate prediction (≥81 percent). This allows drug developers to generate adaptive translational hypotheses to accelerate the development and clinical testing of novel therapeutics for treating a variety of diseases and infections such as cancer, diabetes, and, more recently, COVID-19.
The novel computational framework for designing rule-based simulations of in vivo phenomena includes a database architecture that integrates readily with curated data, tracks subject-specific physiological differences, and enables training and evaluation of model performance using experimental data points across diverse validation datasets.
We use world-class expertise in machine learning, mechanistic modeling, and engineering to produce models that serve as ‘digital’ representations of animals, including humans
BIOiSIM includes a robust model for predicting both small and large molecules in vivo pharmacokinetics, as well as for healthy and patient biomarker pathways. Because the model has been developed and assessed against known outcomes, the complexity of the underlying core model equations is balanced with the ability to translate to new experimental conditions by a critical analysis of model performance. The platform has been applied successfully to the optimization of transdermal dosing for combination therapies, modeling of diabetes signaling pathways, and is currently being expanded from modalities of small molecules to large molecules, CRISPR-Cas, and viral systems and drug combinations against COVID-19.
How does VeriSIM’s platform enable faster model development, more accurate prediction, and higher scalability? Please elaborate.
BIOiSIM is a dynamic, biology-driven platform that provides a scalable computational prediction of in vivo PK/PD phenomena. A whole-body 16-compartment model of systems biology and PK/PD is integrated with AI/ML to make accurate and computationally scalable predictions that are applicable to compound datasets with gaps and variability. The integration of ML with mechanistic modeling allows the platform to fill-in missing data gaps through optimization/prediction given datasets of various completeness.
We have invested heavily in our database and simulation capabilities. VSL has developed the largest proprietary curated database consisting of structure-related data for >1M compounds, >3700+ unique in vivo plasma concentration-time validation datasets from public and proprietary sources representing ~2000 unique compounds and 83 different subject populations (different species, gender, strain, sub-strain). Further, our software systems provide the simulation time for a drug-species PK/PD experimental setup in the order of seconds; this scalability escalates computational capabilities beyond the current paradigm of on-premise pharmacological software.
BIOiSIM can predict PK-specific drug toxicities based on various drug dosages and methods of the drug’s administration (oral, IV, intranasal, or transdermal) in healthy human/animal subjects with over 81 percent accuracy, validated by existing preclinical and clinical data.
By integrating machine learning algorithm-based approaches into traditional physiological modeling and simulation, we can build more complicated physiologically-based models informed by different cell types, pathways, and signaling components, to ultimately reduce drug failure rates in clinical trials and minimize the need for animal testing. We also use a multi-omics approach that integrates how personalized, patient-specific, drug-specific, and disease-specific parameters affect how the body responds to a drug and how a drug responds to the body. The result is that VSL can provide more precise data to facilitate decision making during drug development.
What sets VSL apart from its competitors?
Currently, there are more than 250 companies active that specifically implement AI/ML techniques across the drug life cycle. Most of these companies operate in the early drug discovery phase (high throughput target screening) instead of the late discovery phase and early drug development where VSL services are being provided. Moreover, we are complementary to several of these AI/ML companies in the drug discovery space.
Our direct competitors are mechanistic modeling companies, but our customer-validated value propositions and unique ML/AI integrations scale beyond their offerings into formulation strategy, experimental drug designs, and patient stratifications. This gives us an additional competitive edge in the market. Further, we have established unique multi-way partnerships with industry and academia to develop joint-IPs in addition to providing software subscription, which not only expands our assets but also leads in capturing the market even further. The holistic integration of various datasets makes us the first-in-business to enable accurate insight and translatability into personalized medicine by integrating species/patient-specific physiological variabilities and disease states into the platform and validating the prediction accuracy across different classes of therapeutics.
What does the future hold for VSL?
Our mission is to help design better drugs, predict disease outcomes, and select specific patient populations who would benefit from therapies, making the dream of personalized medicine a reality. For this, we will grow a team to include world-class talent in the engineering and operations space and keep optimizing our modeling and simulation practices through expanded partnerships. As we expand, we will maintain a cross-collaborative culture reliant upon world-class subject matter expertise and an approach to collaboration that seeks to include as many diverse experiences as possible.