In the pharmaceutical industry, the distance between a successful in vitro result and a successful in vivo outcome is a gap paved with billion-dollar losses and abandoned breakthroughs.

Recent research published in Nature Reviews Drug Discovery suggests that while technical success in early-stage discovery has increased, the translational success rate remains stubbornly low. In fact, roughly 90% of drug candidates fail during clinical trials, with a staggering percentage of those failures attributed to poor bioavailability and toxicity issues that should have been predicted early through in vitro–in vivo correlation (IVIVC).

The problem is not a lack of data, it’s the lack of meaningful correlation between what is observed in vitro and what actually happens in vivo.

The Problem: Weak IVIVC and the Cost of Disconnected Data

Most discovery programs suffer from a data silo problem. While teams generate massive amounts of data, the insights are often fragmented:

• In vitro screening results lacked predictive insight into expected in vivo performance allowing high-risk compounds to progress as false positives.
• Translational risk became visible only after animal studies.
• Historical in vitro and in vivo datasets remained disconnected across programs, limiting cross-study learning.
• When a compound fails late in the game, the CRO-Sponsor relationship is tested.

At its core, the issue is simple: IVIVC is either weak, delayed, or entirely missing in most discovery workflows.

The Solution: A Translation-Aware Data Foundation

Improving drug success rates requires strengthening IVIVC. This can be enabled by building a translation-aware data foundation that connects molecular structure, assay signals, and in vivo outcomes into a unified system. The solution involves:

• Predictive IVIVC Modeling: Estimating in vivo outcomes at the point of in vitro data generation
• Cross-Program Learning: Using historical datasets to continuously refine IVIVC accuracy
• Decision-Ready Outputs: Translating complex data into clear Go/No-Go signals

Real World Impact :

Elucidata partnered with a global CRO supporting pharmaceutical discovery programs across in vitro screening, ADME, and animal PK–PD studies. Despite generating strong in vitro leads, the organization faced a recurring challenge i.e. many compounds failed to translate effectively in vivo, particularly in biodistribution studies. This gap created uncertainty for sponsors and raised concerns about the predictive reliability of their in vitro ADME assays.

By systematically analyzing historical study data and integrating cross-stage insights, we identified key disconnects between in vitro readouts and in vivo outcomes. Specifically, we uncovered patterns in compound properties and assay conditions that were consistently associated with downstream failures but were not being captured in early-stage evaluations.

As a result, the CRO was able to refine its screening framework- introducing more predictive markers, improving assay calibration, and prioritizing compounds with higher translational potential. This led to:

• Earlier identification of translational risk during compound selection
• Improved confidence in in vitro–in vivo correlation
• Reduced costly re-evaluation cycles and unnecessary animal studies
• Stronger, data-backed communication with sponsors

In a competitive CRO landscape, translational confidence is a premium service. By adopting an IVIVC-driven prediction model, you can move faster, spend smarter, and bring life-saving treatments to patients with a level of certainty that was previously impossible.

This work represents a step toward a larger shift: a predictive platform that reads molecular structure first and advances only high-confidence candidates to the lab.

‍