Mogam Institute for Biomedical Research Publishes Paper at World’s Leading AI Conference 'NeurIPS'

Proposal of a New Conformal Prediction Model
Minimizing AI Prediction Uncertainty

The Mogam Institute for Biomedical Research announced on December 11 that a paper on a "conformal prediction framework optimized for label shift" has been published at the 2025 Reliable ML from Unreliable Data workshop of NeurIPS, the world's most prestigious artificial intelligence conference.

This research introduces a model that predicts AI uncertainty through a new conformal prediction technology in AI-based drug discovery.

Drug development has long been considered a major challenge due to its enormous costs and high failure rates. In particular, accurately predicting molecular properties such as solubility, efficacy, and toxicity of candidate compounds is a crucial step in determining clinical viability. While artificial intelligence has recently accelerated this process, the issue of "distribution shift"-where predictive performance drops when experimental conditions differ from training data-remains a significant obstacle to reliability. Additionally, conventional AI models only provide single-value predictions, which limits decision-making in experimental design stages where uncertainty is high.

The Mogam Institute research team has proposed a conformal prediction framework optimized for label shift. This technology statistically adjusts prediction intervals to account for changes in label distribution without retraining the data, and provides confidence intervals for each prediction.

Notably, the approach proposed by the Mogam Institute quantitatively presents the uncertainty of AI predictions, thereby helping to mitigate experimental uncertainty-one of the greatest hurdles in drug development. It also meets the transparency and reliability standards required by regulatory agencies, and is expected to further accelerate the use of AI in real-world pharmaceutical development pipelines.

Shin Hyunjin, Director of the Mogam Institute for Biomedical Research, stated, "Having our paper published at a world-renowned AI conference is recognition of our institute's research achievements," adding, "The Mogam Institute for Biomedical Research will continue to fulfill its role as a leading institution in AI-driven drug discovery."

The Mogam Institute for Biomedical Research is a non-profit research foundation established by GC Green Cross. Its research areas span basic biotechnology, vaccine and therapeutic development, and, more recently, AI-based drug discovery.

This content was produced with the assistance of AI translation services.