While large language models (LLMs) are rapidly advancing scientific research, they continue to struggle with core biological reasoning tasks essential for translational and biomedical discovery. To address this limitation, we created and curated eight comprehensive benchmark datasets comprising over 300,000 verifiable question-and-answer pairs, each targeting critical challenges in drug discovery including target druggability, modality suitability, and drug perturbation effects. Using this resource, we developed the OwkinZero models by post-training open-source LLMs through a Reinforcement Learning from Verifiable Rewards strategy. Our results demonstrate that specialized 8-32B OwkinZero models substantially outperform larger, state-of-the-art commercial LLMs on these biological benchmarks. Remarkably, we uncover evidence of a key aspect of generalization: specialist models trained on a single task consistently outperform their base models on previously unseen tasks. This generalization effect is further amplified in our comprehensive OwkinZero models, which were trained on a mixture of datasets and achieve even broader cross-task improvements. This study represents a significant step toward addressing the biological reasoning blind spot in current LLMs, demonstrating that targeted reinforcement learning on carefully curated data can unlock generalizable performance in specialized models, thereby accelerating AI-driven biological discovery.

For this study, Lindsey and his colleagues worked to lay down some of that groundwork. Previous research has shown that various dimensions of LLMs' behavior--from whether they are talking about weddings to persistent traits such as sycophancy--are associated with specific patterns of activity in the simulated neurons that constitute LLMs. Those patterns can be written down as a long string of numbers, in which each number represents how active a specific neuron is when the model is expressing that behavior. Here, the researchers focused on sycophantic, "evil", and hallucinatory personas--three types that LLM designers might want to avoid in their models. To identify those patterns, the team devised a fully automated pipeline that can map out that pattern given a brief text description of a persona.

Machine learning has been applied to a wide variety of genomic prediction problems, such as predicting transcription factor binding, identifying active cis-regulatory elements, constructing gene regulatory networks, and predicting the effects of single nucleotide polymorphisms. The inputs to these models typically include some combination of nucleotide sequence and signals from epigenomics assays. Given such data, the most common approach to evaluating predictive models is a "cross-chromosomal" strategy, which involves training a separate model for each cell type and partitioning genomic loci into some number of folds for cross-validation (Figure 1a). Typically, the genomic loci are split by chromosome. This strategy has been employed for models that predict gene expression [1–3], elements of chromatin architecture [4, 5], transcription factor binding [6, 7], and cis-regulatory elements [8–13]. Although the cross-chromosomal approach measures how well the model generalizes to new genomic loci, it does not measure how well the model generalizes to new cell types.