AI drug development moves towards virtual human bodies: Differentia Bio uses prospective clinical evidence to verify the new paradigm

"AI drug development is currently stuck at a threshold known as the "Phase II Chasm": while molecular optimization capabilities have advanced by leaps and bounds, the prediction of human therapeutic efficacy still relies on trial and error. A company named Differentia Bio has provided a sample worthy of the industry's re-examination with a three-year-long chain of evidence."

A widely recognized industry dilemma: AI can design molecules, but cannot accurately predict their performance in human bodies

Over the past decade, the main battlefield of AI in drug R&D has been the preclinical stage. The efficiency of links such as molecular generation, ADMET prediction, and target binding affinity optimization has been significantly improved. But a more critical problem is emerging.

A 2024 industry analysis by Boston Consulting Group presented two sets of key data: drugs discovered by AI have a success rate of 80%—90% in Phase I clinical trials, far higher than the historical level of about 52% for traditional drugs; but after entering Phase II, the success rate plummets to about 40%, which is almost on par with the roughly 37% success rate of traditional drugs.

This means that AI excels at "producing molecules" rather than "predicting whether a molecule is truly effective in complex human diseases". The biological differences between animals and humans, and the high heterogeneity of patient disease mechanisms, always bring huge uncertainty to clinical translation. What is more tricky is that after entering the clinical stage, early data are usually sparse and fragmented, which can only reflect the results under limited patients, doses and treatment regimens. Faced with problems that directly affect the success or failure of R&D, such as "what will happen if the dose is adjusted", "what will happen if the patient group is changed" and "what will happen in combination therapy", pharmaceutical companies often have to make high-cost decisions based on limited data.

This is exactly the proposition for the second half of AI-driven pharmaceutical R&D: AI not only needs to be able to generate molecules, but also needs to understand complex human diseases, simulate drug interventions, and provide scientific basis for R&D decisions before clinical results are obtained. Shifting from molecular generation to human disease simulation is becoming a noteworthy new direction for AI-powered drug R&D.

A rare chain of evidence: prediction comes first, clinical verification follows

Against this background, a forward-looking verification progress announced by Differentia Bio, an AI digital twin enterprise headquartered in San Francisco, USA, provides a case worthy of attention.

The company's prediction of the therapeutic mechanism for IgG4-related disease (IgG4-RD) has obtained important support from subsequent independent human clinical studies and FDA-approved drugs.

2023—2024: Prospective model predictions and public academic records. The Differentia team integrated patient multi-omics data, AI knowledge graph and quantitative disease mechanism model, systematically simulated and prioritized potential therapeutic pathways for IgG4-RD, and identified intervention directions with therapeutic potential such as CD19+ and IL-4/Th2. Relevant achievements were publicly displayed at international academic conferences such as ACOP, PAGE and NBC, leaving traceable early scientific records.

April 2025: CD19+ targeted therapy is approved by the FDA. CD19+ targeted inebilizumab was approved by the US FDA for IgG4-RD, becoming the first approved therapeutic drug for this indication. Its pivotal Phase III clinical study showed that within the 52-week treatment period, compared with placebo, the risk of disease flare was reduced by 87% (HR=0.13, p<0.0001).

January 2026: The IL-4/Th2 pathway receives further independent clinical support. A third-party research team published an early clinical study of the IL-4Rα targeted drug dupilumab for the treatment of IgG4-RD: 4 out of 5 patients showed clinical improvement and decreased serum IgG4 levels, and 3 of them had IgG4 levels restored to the normal range.

For complex diseases, the difficulty of drug R&D is often not the lack of potential targets, but how to judge which intervention mechanisms are most worthy of investment in clinical development when facing dozens of therapeutic hypotheses and scattered, even contradictory scientific evidence.

The core capability of Differentia is not simply discovering brand new targets, but simulating the intervention effects of different therapeutic pathways under a unified computing framework, so as to systematically compare and prioritize candidate therapeutic strategies.

In the AI drug R&D industry, a large number of platforms demonstrate their predictive capabilities through historical data backtesting, retrospective analysis, or cellular and animal experiment results. The difference of the Differentia case lies in that model predictions and public academic records come first, and independent human clinical results provide support afterwards. This provides a rare reference for evaluating the predictive ability of AI disease models — instead of "fitting the past", it is "predicting the future and being verified".

More importantly, the scarce prospective clinical verification is building a long-term competitive barrier for Differentia that is difficult to replicate. Algorithms and models can be iterated continuously, but the scientific evidence with prediction first and independent clinical verification afterwards requires years of real human research and time accumulation, which cannot be rapidly replicated through historical backtesting or massive capital investment. With the continuous accumulation of more independent verifications across diseases and therapeutic mechanisms, this traceable predictive capability is expected to become a continuously expanding technical and trust advantage for Differentia.

World-class industrial and scientific leadership team builds interdisciplinary technical barriers

Differentia brings together a core scientific team with both top international academic background and R&D experience of the world's top MNCs, including Catherine Sherwin, currently an adjunct professor of clinical pharmacology at the University of Minnesota, USA; Atefeh Kazeroonian, former PI of computational immunology at Technical University of Munich, Germany; Simon Junankar, former PI of tumor immunology at the University of New South Wales (UNSW), Australia; and Kas Subramanian, the global QSP scientific leader with more than 30 years of industry experience, former executive of modeling R&D at Applied BioMath, and also the CEO of the company.

Different from many start-ups that mainly rely on external academic consultants or part-time experts, Differentia deeply integrates multi-domain scientific leadership into its core R&D system, integrating key fields such as artificial intelligence, multi-omics bioinformatics, computational immunology, tumor biology, quantitative systems pharmacology, clinical pharmacology, and regulatory science. This allows Differentia to closely combine cutting-edge AI, real patient multi-omics data, in-depth disease biology and quantitative mechanism simulation, establish complete capabilities from scientific hypotheses to clinical decisions, and lay a solid foundation for building digital twins of complex human diseases.

Why this matters to the industry: from "black box correlation" to "computable mechanism"

While technologies such as virtual cells and virtual tissues continue to expand the boundary of AI simulating living systems, Differentia has extended its simulation capabilities to the level of human diseases and patients, and demonstrated the potential of moving from biological simulation to real drug R&D decisions through prospective predictions and subsequent independent clinical evidence.

To understand the value of this chain of evidence, we need to see the difference between Differentia's technical path and the mainstream AI drug discovery technical path.

Differentia's entry point is not starting from the molecular end, but starting from real patient disease biology. Real patient multi-omics data is an important foundation of this technical system. Relying on its long-established global clinical research cooperation network, Differentia continuously accumulates multi-dimensional multi-omics data from real patients. These multi-level human data provide an important basis for reconstructing complex disease mechanisms, characterizing patient heterogeneity and calibrating dynamic disease models, and also constitute the company's continuously accumulated data and model barriers.

Relying on its self-developed DHARMA™ AI disease digital twin engine, Differentia deeply integrates generative AI, multi-omics machine learning, knowledge graph, systems biology and quantitative mechanism model to build a multi-scale human disease simulation system covering the molecular, cellular, tissue and patient levels. At present, the platform has covered oncology, autoimmune and metabolic diseases, and has the simulation capability for multiple treatment modalities such as small molecules, macromolecules, ADCs, radiopharmaceuticals and CGT.

In this architecture, AI is not only used for knowledge retrieval and data analysis, but also deeply involved in disease mechanism reasoning, model construction and parameter calibration, and orchestrates virtual experiments, explores therapeutic hypotheses and analyzes simulation results through AI Agents. The underlying quantitative mechanism model provides a dynamic computing environment that conforms to biological laws, so that scientific hypotheses generated by AI can be tested through interpretable mechanism simulation.

This is not simply using a large model to predict whether a drug will succeed, but making generative AI work collaboratively with computable human disease models to form a closed loop from scientific reasoning, virtual experiments to R&D decisions.

This capability mainly creates value in three key links of drug R&D:

· Preclinical stage: Improve the quality of candidate drug selection. By simulating the potential impact of different therapeutic mechanisms on human diseases, high-potential therapeutic directions can be identified earlier, and projects destined to fail in human trials can be prevented from entering the clinical stage.

· Clinical development: Optimize trial design and reduce avoidable failures. Although only hundreds of real clinical trials are conducted for one pipeline, combining more than 2 million virtual patient simulations to systematically compare more than 5000 different scenarios of doses, patient stratification, indications, combination therapies and competing products can maximize the success rate of clinical trials.

· Phased decision-making: Improve the efficiency of R&D resource allocation. With the continuous accumulation of clinical data and continuous calibration and update of models, it helps the R&D team analyze the potential causes of poor efficacy, distinguish mechanism limitations from optimizable development problems, and support key decisions such as continuing development, adjusting strategies or terminating projects.

Differentia is integrating AI-driven human disease simulation into key decision-making links from candidate drug selection to clinical development. Drug R&D is not short of scientific hypotheses, but lacks the ability to systematically compare, quantitatively evaluate and prioritize the most promising therapeutic strategies in complex human diseases.

The regulatory window is opening: Why now

The development of human disease digital twins is ushering in an important window period for the evolution of global regulatory methodology.

Major regulatory agencies such as the FDA and EMA continue to promote computational modeling and Model-Informed Drug Development (MIDD). In 2026, ICH M15 "General Principles for Model-Informed Drug Development" officially enters the implementation stage, establishing a more coordinated international framework for the planning, evaluation, reporting and regulatory communication of model evidence. China's Center for Drug Evaluation (CDE) also continuously promotes the technical guidelines and regulatory practices related to model-informed drug R&D.

Regulators are establishing a clearer evaluation path for computational models to participate in drug R&D, and the scientific credibility, applicability and evidence quality of models will become the key.

Differentia's digital twin models are adaptable in these three dimensions — all prediction results are derived based on real human disease mechanisms, with complete academic records and clinical verification basis, and can directly serve compliant R&D and clinical declaration work.

For the industry, this means that "virtual models participating in drug R&D" is no longer a long-term concept, but is entering a window period of compliance and normalization. Whoever accumulates traceable prospective verification records in this window period will have the first-mover advantage.

From verification to commercialization: Embedding in core decision-making links

At the commercial level, the company has carried out commercial cooperation with many large global multinational pharmaceutical enterprises (MNCs), leading Biotechs, clinical CROs and investment institutions, applying AI-driven digital twin simulation to key scenarios such as therapeutic mechanism evaluation, R&D strategy optimization, clinical trial design and asset value assessment.

It is worth noting that Differentia has achieved profitability through simulation revenue. In the AI life science field with long R&D cycle and high capital investment, this further verifies the actual commercial value of the company's technology.

From science fiction imagination to clinical verification: A new paradigm of drug R&D is opening

Allowing AI to simulate complex human diseases and predict the potential efficacy of different drugs before real patients receive treatment was once considered a distant scientific vision.

Now, with a chain of prospective predictions and independent clinical evidence spanning several years, Differentia is making this vision move from concept to reality. Behind this points to an underlying efficiency transformation that reshapes drug R&D — systematically reducing avoidable clinical failures, greatly increasing R&D success rates, shortening R&D cycles, and achieving revolutionary improvements in R&D efficiency and return on investment.

Its far-reaching significance is that it enables humans to gradually acquire the ability to understand, simulate and predict complex diseases, so as to accelerate the deciphering of disease mechanisms and find more treatment solutions that were difficult to achieve in the past.

Rehearse in the virtual human body, verify in real clinical practice. The paradigm revolution that changes the economics of drug R&D and even the way humans understand and treat diseases has kicked off.