Teodoro Laino

Heterogeneous catalyst design is primarily expert-driven and difficult to keep track amidst rapidly growing literature, motivating the use of large language models (LLMs) to learn from existing knowledge. While generalist LLMs show some capability in generating synthesis procedures, they struggle on encountering complex catalytic systems like single-atom catalysts (SAC). Here, we investigate fine-tuning of LLMs to improve the generation of SAC synthesis procedures with chemical relevance. We curate a dataset of 2,964 SAC publications and train the Granite-based LLM to recommend protocols based on user-defined prompts including metal-support combinations, synthesis method, and target reactions. We demonstrate the model’s practicality through a user interface allowing researchers to query procedures tailored to their design conditions. Human-in-the-loop validation of 150 model-generated SAC protocols confirmed coherent, context-aware procedures with synthesis sequence steps that align with experimental logic, along with limitations. We underscore LLM-generated protocols must be critically reviewed and used as assistive tools, and highlight fine-tuned LLMs for AI-assisted synthesis planning driving innovation in catalysis research.

Peptide aggregation is a long-standing challenge in chemical peptide synthesis, limiting its efficiency and reliability. Although data-driven methods have enhanced our understanding of many sequence-based phenomena, no comprehensive approach addresses so-called non-random difficult couplings (generally linked to aggregation) during solid-phase peptide synthesis. Here we leverage existing peptide synthesis datasets, supplemented with further experimental data, to build a predictive model that deciphers the role of individual amino acids in triggering aggregation. We first identified and experimentally validated composition-dependent aggregation as a stronger predictor than sequence-based patterns. This insight enabled the development of a composition vector representation, allowing insights into the aggregation propensities of individual amino acids. Applying an ensemble of trained models, we predicted the aggregation properties of peptides and recommended the optimized use of aggregation-reducing tools. By elucidating each individual amino acid’s influence, this method holds the potential to accelerate synthesis optimization through existing data, offering a robust framework for understanding and controlling peptide aggregation.

Bridging AI and self-driving laboratories, we introduce the first fully-automated, closed-loop molecular discovery cycle, exemplified by the identification of novel JAK inhibitors. With minimal human intervention, we combined AI-driven molecular design and retrosynthesis with IBM’s synthesis automation system RoboRXN and Arctoris’ Ulysses platform for automated in-vitro screening. We performed two Design-Make-Test-Analyze (DMTA) cycles with a total of 36 synthesized compounds. In the first round, the generative model identified structural analogs of established JAK inhibitors despite being blinded from known JAK inhibitors. The second iteration yielded molecules with significantly improved pIC50 and ligand efficiency (p < 0.001) compared to the first round, underscoring the effectiveness of the closed-loop workflow.

Combining AI and self-driving laboratories unlocks closed-loop molecular discovery by automation across all stages of the Design-Make-Test-Analyze (DMTA) cycle.

Millions of synthesis procedures have been published, but their meta-analysis (e.g., to identify successful synthesis patterns) is troublesome due to the variety of reporting structures. Mapping unstructured language to a precise sequence of actions requires an understanding of domain-specific jargon, a challenge usually addressed by fine-tuning models on labeled data. Herein, we present a simple, training-free workflow for laboratory action extraction that works across fields. It encodes domain knowledge through explicit action sets and uses powerful, readily available Large Language Models (LLMs). Applied to zeolite synthesis, our approach outperforms existing open-source tools and delivers context-aware results with open, locally runnable LLMs. It also matches the performance of state-of-the-art, field-specific models in their own domains, highlighting the generalization ability of current LLMs. With our methodology and open algorithms, chemists can evaluate actions sets, screen LLMs for their specific needs, and accurately digitize laboratory procedures with minimal effort.

Phosphorus-31 nuclear magnetic resonance (³¹P NMR) spectroscopy is a powerful technique for characterizing phosphorus-containing compounds in diverse chemical environments. However, spectral interpretation remains a time-consuming and expertise-dependent task, relying on reference tables and empirical comparisons. In this study, we introduce a data-driven approach that automates ³¹P NMR spectral analysis, providing rapid and accurate predictions of the local phosphorus environments. By leveraging a curated data set of experimental and synthetic spectra, our model achieves a Top-1 accuracy of 53.64% and a Top-5 accuracy of 77.69% at predicting the local environment around a phosphorus atom. Furthermore, it demonstrates robustness across different solvent conditions and outperforms expert chemists by 25% in spectral assignment tasks. The models, data sets, and architecture are openly available, facilitating seamless adoption in chemical laboratories engaged in structure elucidation, with the goal of advancing ³¹P NMR spectral analysis and interpretation.

Transformer language models recently enabled molecular structure prediction directly from infrared (IR) spectra, yet have remained confined to pure compounds. We show that the same architecture learns the correlations embedded in binary mixture spectra and can retrieve the individual molecular components. Trained solely on gas-phase data, our model attains a Top-10 accuracy of 61.4% on balanced synthetic mixtures. When evaluated on 15 mixtures measured with Attenuated Total Reflectance (ATR) IR spectrometer, whose response differs markedly from the training domain, it still achieves 52.0% Top-10 accuracy, evidencing strong cross-instrument transferability. The ability to identify signals of individual molecules within complex spectra extends machine-learning-assisted spectroscopy from idealised samples to realistic laboratory scenarios. All code and pretrained weights are released to accelerate adoption and further development. This advance opens the door to automated structure elucidation using IR data in fields ranging from environmental monitoring to pharmaceutical quality control.

Chemists can typically only identify certain functional groups from IR spectra. In this work we leverage a transformer model to directly predict the complete molecular structure from a spectrum achieving an accuracy up to 64%