Julien Fauqueur

Josefa Lia Stoisser, Marc Boubnovski Martell, Lawrence Phillips et al.

We propose STRuCT-LLM, a unified framework for training large language models (LLMs) to perform structured reasoning over both relational and graph-structured data. Our approach jointly optimizes Text-to-SQL and Text-to-Cypher tasks using reinforcement learning (RL) combined with Chain-of-Thought (CoT) supervision. To support fine-grained optimization in graph-based parsing, we introduce a topology-aware reward function based on graph edit distance. Unlike prior work that treats relational and graph formalisms in isolation, STRuCT-LLM leverages shared abstractions between SQL and Cypher to induce cross-formalism transfer, enabling SQL training to improve Cypher performance and vice versa - even without shared schemas. Our largest model (QwQ-32B) achieves substantial relative improvements across tasks: on semantic parsing, Spider improves by 13.5\% and Text2Cypher by 73.1\%. The model also demonstrates strong zero-shot generalization, improving performance on downstream tabular QA (TableBench: 8.5\%) and knowledge graph QA (CR-LT-KGQA: 1.7\%) without any QA-specific supervision. These results demonstrate both the effectiveness of executable queries as scaffolds for structured reasoning and the synergistic benefits of jointly training on SQL and Cypher (code available at https://github.com/bouv/STRuCT-LLM).

Josefa Lia Stoisser, Marc Boubnovski Martell, Julien Fauqueur

This work reframes the Text-to-SQL task as a pathway for teaching large language models (LLMs) to reason over and manipulate tabular data--moving beyond the traditional focus on query generation. We propose a two-stage framework that leverages SQL supervision to develop transferable table reasoning capabilities. First, we synthesize detailed chain-of-thought (CoT) traces from real-world SQL queries, providing step-by-step, clause-level supervision that teaches the model how to traverse, filter, and aggregate table fields. Second, we introduce a Group Relative Policy Optimization (GRPO) reinforcement learning objective that connects SQL execution accuracy to generalizable reasoning by encouraging steps that extend beyond task-specific syntax and transfer across datasets. Empirically, our approach improves performance on standard Text-to-SQL benchmarks and achieves substantial gains on reasoning-intensive datasets such as BIRD and CRT-QA, demonstrating enhanced generalization and interpretability. Specifically, the distilled-quantized LLaMA model achieved a relative 33.9\% increase in accuracy when trained on Text-to-SQL tasks, while Qwen achieved a relative 14.5\% increase. These results suggest that SQL can serve not only as a target formalism but also as an effective scaffold for learning robust, transferable reasoning over structured data.

Josefa Lia Stoisser, Marc Boubnovski Martell, Lawrence Phillips et al.

Large language model (LLM) agents are increasingly deployed in structured biomedical data environments, yet they often produce fluent but overconfident outputs when reasoning over complex multi-table data. We introduce an uncertainty-aware agent for query-conditioned multi-table summarization that leverages two complementary signals: (i) retrieval uncertainty--entropy over multiple table-selection rollouts--and (ii) summary uncertainty--combining self-consistency and perplexity. Summary uncertainty is incorporated into reinforcement learning (RL) with Group Relative Policy Optimization (GRPO), while both retrieval and summary uncertainty guide inference-time filtering and support the construction of higher-quality synthetic datasets. On multi-omics benchmarks, our approach improves factuality and calibration, nearly tripling correct and useful claims per summary (3.0\(\rightarrow\)8.4 internal; 3.6\(\rightarrow\)9.9 cancer multi-omics) and substantially improving downstream survival prediction (C-index 0.32\(\rightarrow\)0.63). These results demonstrate that uncertainty can serve as a control signal--enabling agents to abstain, communicate confidence, and become more reliable tools for complex structured-data environments.

Josefa Lia Stoisser, Lawrence Phillips, Aditya Misra et al.

Synthetic chain-of-thought (CoT) traces are widely used to train large reasoning models (LRMs), improving generalization by providing step-level supervision. Yet most approaches require ground-truth labels to seed or filter these traces - an expensive bottleneck in domains like biology where wet-lab data are scarce. We propose a label-free alternative: uncertainty-based filtering, which uses a model's own confidence - quantified through established uncertainty metrics like self-consistency and predictive perplexity - as a substitute for external labels. We sample multiple reasoning traces and retain only low-uncertainty subsets. Applied to biological perturbation prediction, a domain where wet-lab labels are especially costly, we show that the filtered subset has higher accuracy, and that supervised fine-tuning (SFT) on uncertainty-filtered data outperforms unfiltered synthetic data, narrows the gap to ground-truth training, and surpasses strong LRM baselines. Ablations show that per-class filtering corrects for class-specific uncertainty scales and that hybrid uncertainty metrics yield higher-quality datasets. Our results suggest that model-internal confidence is a powerful signal for efficient reasoning dataset creation, enabling LRMs in domains where supervision is expensive.

Kalyan Ramakrishnan, Jonathan G. Hedley, Sisi Qu et al.

We train a neural network to predict distributional responses in gene expression following genetic perturbations. This is an essential task in early-stage drug discovery, where such responses can offer insights into gene function and inform target identification. Existing methods only predict changes in the mean expression, overlooking stochasticity inherent in single-cell data. In contrast, we offer a more realistic view of cellular responses by modeling expression distributions. Our model predicts gene-level histograms conditioned on perturbations and outperforms baselines in capturing higher-order statistics, such as variance, skewness, and kurtosis, at a fraction of the training cost. To generalize to unseen perturbations, we incorporate prior knowledge via gene embeddings from large language models (LLMs). While modeling a richer output space, the method remains competitive in predicting mean expression changes. This work offers a practical step towards more expressive and biologically informative models of perturbation effects.

Harshil Shah, Julien Fauqueur

Extracting biomedical relations from large corpora of scientific documents is a challenging natural language processing task. Existing approaches usually focus on identifying a relation either in a single sentence (mention-level) or across an entire corpus (pair-level). In both cases, recent methods have achieved strong results by learning a point estimate to represent the relation; this is then used as the input to a relation classifier. However, the relation expressed in text between a pair of biomedical entities is often more complex than can be captured by a point estimate. To address this issue, we propose a latent variable model with an arbitrarily flexible distribution to represent the relation between an entity pair. Additionally, our model provides a unified architecture for both mention-level and pair-level relation extraction. We demonstrate that our model achieves results competitive with strong baselines for both tasks while having fewer parameters and being significantly faster to train. We make our code publicly available.

Julien Fauqueur, Ashok Thillaisundaram, Theodosia Togia

Knowledge base construction is crucial for summarising, understanding and inferring relationships between biomedical entities. However, for many practical applications such as drug discovery, the scarcity of relevant facts (e.g. gene X is therapeutic target for disease Y) severely limits a domain expert's ability to create a usable knowledge base, either directly or by training a relation extraction model. In this paper, we present a simple and effective method of extracting new facts with a pre-specified binary relationship type from the biomedical literature, without requiring any training data or hand-crafted rules. Our system discovers, ranks and presents the most salient patterns to domain experts in an interpretable form. By marking patterns as compatible with the desired relationship type, experts indirectly batch-annotate candidate pairs whose relationship is expressed with such patterns in the literature. Even with a complete absence of seed data, experts are able to discover thousands of high-quality pairs with the desired relationship within minutes. When a small number of relevant pairs do exist - even when their relationship is more general (e.g. gene X is biologically associated with disease Y) than the relationship of interest - our system leverages them in order to i) learn a better ranking of the patterns to be annotated or ii) generate weakly labelled pairs in a fully automated manner. We evaluate our method both intrinsically and via a downstream knowledge base completion task, and show that it is an effective way of constructing knowledge bases when few or no relevant facts are already available.