Drug development has always been a long, expensive, and uncertain process. Despite advances in biology and chemistry, many promising candidates still fail once they reach human testing. That high attrition rate has fueled interest in new tools that can improve predictions earlier, when decisions matter most.

Artificial intelligence (AI) and mechanistic modeling are no longer fringe tools in drug development; they are rapidly becoming central to how new therapies are discovered, designed, and de-risked. However, adoption is not without challenges. Fragmented data, cultural resistance, and evolving regulatory expectations all continue to shape how quickly these approaches can scale.

To better understand where these technologies are making the biggest impact and where adoption is headed, DDN spoke with Jo Varshney, Founder and CEO of VeriSIM Life.

From your perspective, where is AI and mechanistic modeling currently making the biggest impact in drug development — and are there particular therapeutic areas, molecule types, or development phases where these approaches have proven especially valuable?

AI and mechanistic modeling are making the greatest impact in the early stages of development, particularly from preclinical through Phase 2a.

These approaches are enabling more accurate predictions of toxicity, absorption, distribution, metabolism, and excretion (ADME) properties, and efficacy, while also supporting the creation of virtual populations that capture demographic and physiological variability.

In clinical development, they’re increasingly being used to optimize dose selection, stratify patients, and reduce trial sizes — capabilities that are especially valuable in rare diseases where patient pools are small.

Therapeutic areas seeing the most benefit include oncology, where models inform combination regimens and ADC development; pulmonary and cardiovascular disease, where remodeling, fibrosis, and hemodynamics can be mechanistically simulated; and neurology, where blood-brain barrier permeability and network-level effects are difficult to capture otherwise.

The approaches are well-suited for small molecules, but applications are growing for biologics and complex combinations. Regulators, including the FDA and the European Medicines Agency (EMA), are increasingly receptive under the New Approach Methodologies framework, and industry is adopting them to inform early go/no-go decisions.

How has industry adoption of AI and modeling evolved over the past few years, and what shifts are you seeing now?

Industry adoption of AI and mechanistic modeling has shifted substantially over the last five years. What began as exploratory pilots has evolved into structured, cross-functional platforms integrated across discovery, development, and regulatory workflows. Regulatory acceptance has also increased: The FDA and EMA now explicitly reference model-informed drug development under the New Approach Methodologies framework, and the recent International Council for Harmonization M15 draft guidance reflects global efforts to standardize validation and reporting.

Current trends show three key scientific shifts:

1. Hybridization: Mechanistic models are increasingly coupled with machine learning for parameter estimation, uncertainty quantification, and extrapolation across diverse populations.
2. Expansion beyond discovery: Applications now extend into clinical trial simulation, dose optimization, patient stratification, manufacturing, and post-market safety surveillance.
3. Emphasis on validation and transparency: As models are incorporated into regulatory submissions, reproducibility, explainability, and formal qualification are becoming essential.

In short, the field has matured from proof-of-concept to a recognized pillar of drug development, with the current inflection point centered on hybrid mechanistic-AI models, regulatory alignment, and extension into later stages of the drug lifecycle.

What are the biggest barriers to broader adoption — technical, cultural, or regulatory?

On the technical side, interoperability remains a core challenge. Biomedical data is still highly fragmented across labs, institutions, and formats. Harmonizing these datasets, ensuring quality, and integrating them into validated models requires significant investment and cross-disciplinary expertise. Validation itself is another hurdle, building confidence that model predictions reliably translate to clinical reality. While substantial progress has been made, the burden of proof remains appropriately high, given the implications for patient safety.

Many scientists were trained to view animal data as the gold standard, so adopting AI- and human-first approaches requires unlearning ingrained habits.
- Jo Varshney

Culturally, the industry is still shifting away from decades of reliance on animal models as the default. Many scientists were trained to view animal data as the gold standard, so adopting AI- and human-first approaches requires unlearning ingrained habits. There is also concern that these technologies could replace rather than empower researchers. Yet in practice, teams that engage with these tools find that they augment expertise and free up scientists to focus on higher-value questions. Scaling that shift in mindset across organizations remains a gradual process.

Regulatorily, we are in a transitional phase. Agencies like the FDA and the National Institutes of Health (NIH) are sending clear signals encouraging non-animal, human-relevant approaches, and international guidelines such as ICH M15 are emerging. But questions persist: Will this data be accepted? Could reliance on new methods slow regulatory timelines? Until standards are fully codified and consistently applied, some companies will hesitate to move beyond traditional approaches.

Saying this, cultural resistance is softening as more scientists experience success stories firsthand, and regulatory frameworks are evolving faster than ever before.

How does BIOiSIM™ differ from other AI platforms?

BIOiSIM® distinguishes itself by both scope and depth. While most AI platforms are designed for narrow use cases such as a single disease, a single target, or limited data inputs, BIOiSIM® spans a wide range of therapeutic areas. It has been applied successfully across pulmonary hypertension, substance use disorders, complex oncological indications, and around 60 other diseases, guiding not only the internal drug assets but also those of our pharmaceutical partners.

At the center is AtlasGEN Novel Drug Designer™, the only generative AI engine that embeds biological data constraints directly into the design process. Instead of producing thousands of theoretical molecules with little clinical relevance, AtlasGEN™ delivers novel compounds that are biologically meaningful, actionable, and ready for testing. This precision accelerates discovery, reduces wasted effort, and enables teams to focus on candidates with true translational potential.

Complementing this is our Translational Index™, the industry’s first AI-based “credit score” for preclinical assets. By quantifying a candidate’s probability of clinical success, the Translational Index™ gives researchers a transparent, data-backed benchmark for decision-making. It helps eliminate low-probability candidates early and prioritize those with the highest chance of success.

Together, AtlasGEN™ and the Translational Index™ make BIOiSIM® a comprehensive, biology-first platform. It not only generates better drug candidates but also gives partners the confidence and clarity to advance them rapidly toward the clinic.

When engaging with industry and regulators, what kinds of evidence or validation have been most effective in building confidence in your models?

What resonates most with both industry and regulators is evidence that goes beyond theory and translates into real-world outcomes. The turning point in these conversations comes when we can show that our models not only replicate established clinical data but also actually anticipate it. For regulators, this means demonstrating concordance between our simulations and observed human dose exposure, or toxicities, often predicting outcomes that were later confirmed in the clinic. That alignment provides an immediate anchor of credibility.

With industry partners, the most powerful proof is prospective validation, where our platform actively shapes decisions in live programs. BIOiSIM® has prevented costly late-stage failures by flagging liabilities early, while also accelerating promising candidates that went on to align with clinical results. These case studies of avoided risk and realized opportunity are what move conversations from curiosity to conviction.

Equally important is transparency and interpretability. We do not hand over “black box” predictions. Every output is tied to mechanistic explanations of how a drug engages with human biology, quantified through our Translational Index™. This gives stakeholders a clear rationale for why a candidate is likely to succeed or fail and makes the results actionable in real decision-making contexts.

Ultimately, confidence is built through a layered approach: retrospective benchmarking that proves accuracy, prospective application that proves utility, and mechanistic transparency that proves explainability. Together, these dimensions demonstrate that our models do more than simulate — they deliver clinically meaningful insights that change the trajectory of development programs.

Could you share an example where your AI platform has significantly reduced time, cost, or reliance on early-stage animal or bench testing?

BIOiSIM® has repeatedly shown it can cut years and millions off development timelines while dramatically reducing reliance on animal testing.

Take drug-induced liver injury (DILI), one of the leading causes of late-stage failures. In a recent project, our hybrid AI model boosted prediction accuracy from just 50 percent with conventional AI to 86 percent — and it did so with limited data. For regulators and industry, this provided a scalable, human-relevant alternative to traditional toxicity screens, reducing the need for early animal or in vitro studies.

In another case, a partner needed novel, clinically relevant compounds designed and ready for preclinical testing in weeks, not months. Within one month, we delivered three synthesis-ready candidates optimized for bioavailability and dosing, resulting in a 50 percent reduction in early development time.

We also applied BIOiSIM end-to-end to advance PT001, an inhaled therapeutic for pulmonary arterial hypertension. By guiding discovery, formulation, and dose optimization in silico, we eliminated off-target risks and moved the program into Investigational New Drug-enabling studies in just two and a half years, nearly twice as fast as industry norms. This program has an FDA Orphan Drug Designation and is positioned to exceed commercial benchmarks.

Across programs, our platform consistently shortens timelines by an average of two and a half years, reduces development costs by more than $3 million per asset, and cuts animal testing by over 75 percent, all while increasing confidence in clinical success.

Do you think a combination of modeling, AI, and advanced organoid systems could eventually replace all animal testing?

The combination of AI, mechanistic modeling, and advanced human-relevant systems such as organoids has the potential to reduce significantly and, in many cases, replace animal testing. It is not about a sudden or total replacement. For decades, animal studies have been the default, and while they have provided some insights, they often fail to predict human outcomes accurately. By contrast, AI-driven models and organoid systems allow us to simulate human biology with greater fidelity, explore multiple hypotheses rapidly, and uncover mechanistic insights that are directly relevant to patients.

I see a future where human-relevant computational and organoid-based approaches become the primary path for drug development, with animal studies reserved only for very specific and scientifically justified circumstances.
- Jo Varshney

This shift is already evident in areas such as neuropharmacology, oncology, and rare diseases, where human-relevant models reveal effects that traditional animal studies miss. The key to fully replacing animal testing lies in validation and regulatory acceptance. We need to consistently and transparently demonstrate that these models can predict clinical outcomes reliably. That is why platforms such as BIOiSIM® are designed not only to generate predictions but also to provide mechanistic explanations and quantitative metrics, such as the Translational Index™, that help regulators and industry build confidence in the results.

I see a future where human-relevant computational and organoid-based approaches become the primary path for drug development, with animal studies reserved only for very specific and scientifically justified circumstances. This is not only a technological shift but also an ethical and scientific evolution toward a more predictive, efficient, and patient-centered model of medicine.

In the next two to three years, where do you expect to see the fastest growth in the use of AI and modeling?

Over the next two to three years, the fastest growth will be in the preclinical and translational stages of drug development, where most attrition and cost occur. Companies are increasingly recognizing that early, human-relevant predictions can dramatically de-risk programs before they enter expensive clinical trials.

We will see greater adoption in target validation, lead optimization, and safety profiling, particularly for complex modalities like biologics, gene and cell therapies, and multi-target compounds, where traditional animal models often fall short.

Another major area of growth will be integrated, human-first platforms that combine AI, mechanistic modeling, and advanced systems such as organoids and microphysiological models. These approaches enable more accurate predictions of efficacy and safety, while also improving clinical trial design and patient stratification. The result is shorter timelines, higher success rates, and reduced reliance on animal testing.

Regulatory momentum is also accelerating this shift. With the FDA encouraging the use of non-animal evidence and the NIH pushing for more human-based research, AI and modeling are moving from pilot projects and niche experiments to central tools in early-stage decision-making.

In short, the next few years will mark a transition from experimentation to adoption, and AI-enabled mechanistic modeling will become the core driver of how drugs are discovered, developed, and de-risked.

This interview has been condensed and edited for clarity.