DrKirill Veselkov

Research Overview:

My group develops translational machine learning and AI methods, including deep and graph neural networks, causality-inspired models, and foundation architectures, to improve personalized medicine, precision nutrition, and population health. We integrate large, heterogeneous datasets spanning multi-omics, mass spectrometry, biomedical imaging, and longitudinal clinical data to quantify disease mechanisms, predict intervention response, and support clinical decision tools.

1) AIDA: Artificially Intelligent Diagnostic Assistant for gastric inflammation. Rising antibiotic resistance has rendered traditional "one-size-fits-all" treatment protocols for H. pylori increasingly ineffective. I led the development of AIDA (Artificially Intelligent Diagnostic Assistant), a first-of-its-kind framework that transforms gastric inflammation management through machine learning. A key component of this work was my development of the "AI-clinician," which utilizes deep reinforcement learning to personalize antibiotic regimens. By harmonizing and analyzing a large dataset of over 75,000 patients, the model identifies optimal treatment pathways that significantly outperform current standard-of-care protocols in eradication success. In addition to clinical decision support, I pioneered the application of foundational AI models to automate the interpretation of pathology and endoscopy imaging. This research addresses the critical issue of diagnostic variability in precancerous gastric conditions. By integrating high-precision computer vision with clinical data, my work provides a scalable, standardized solution for the early detection and precise intervention of inflammatory gastric diseases.

Selected Publications:

a. Higgins, K., Nyssen, O. Southern, J., Laponogov, I., Veselkov, D., Gisbert, J., Fleitas Kanonnikoff, T., & Veselkov, K.* (2025). The Helicobacter pylori AI-clinician harnesses artificial intelligence to personalise H. pyloritreatment recommendations. Nature Communications, 16(1), 6472.

b. Kerdegari, H., Higgins, K., Veselkov, D., Laponogov, I., Polaka, I., Coimbra, M., Pescino, J. A., Leja, M., Dinis-Ribeiro, M., Fleitas Kanonnikoff, T., & Veselkov, K.* (2024). Foundational models for pathology and endoscopy images: Application for gastric inflammation. Diagnostics, 14(17), 1912.

2) MSHub: Scalable AI Framework for Mass Spectrometry and Clinical Breathomics. The lack of standardized, scalable tools for processing complex mass spectrometry data has historically limited the translational utility of metabolomics. I led the development of MSHub, a high-performance machine learning platform designed for the automated deconvolution and annotation of large-scale chromatography-mass spectrometry datasets. Utilizing network-based AI, MSHub is the unique solution capable of handling "big data" cohorts exceeding 10,000 samples with high reproducibility. Now integrated into the Global Natural Products Social Molecular Networking (GNPS) ecosystem, the platform supports a global community of over 1,000 scientists, democratizing advanced computational metabolomics. My team successfully translated this computational framework into the clinical domain, specifically in the field of breathomics. We achieved the first molecular spatial map of the human volatilome, providing a baseline for understanding systemic metabolic signatures. This work directly led to the identification of volatile organic compound (VOC) biomarkers for the non-invasive early detection of esophageal and gastric cancers. By bridging high-performance computing with clinical diagnostics, this research establishes a foundation for scalable, breath-based cancer screening.

Selected Publications:

a. Aksenov AA, Laponogov I, etc, Dorrestein PC, Veselkov K*. Auto-deconvolution and molecular networking of gas chromatography–mass spectrometry data. Nature Biotechnology, 39(2), 169–173. *Corresponding author

b. Abbassi-Ghadi N*, Antonowicz S*, McKenzie J, Kumar S, Huang J, Jones EA, Strittmatter N, Petts G, Kudo H, Court S, Hoare JM, Veselkov K, Goldin R, Tak.ts Z, Hanna GB. De Novo Lipogenesis Alters the Phospholipidome of Esophageal Adenocarcinoma. Cancer Research. 2020 1;80(13):2764-2774.

c. Kumar S, Huang J, Abbassi-Ghadi N, Mackenzie HA, Veselkov KA, Hoare JM, Lovat LB, Španěl P, Smith D, Hanna GB. Mass Spectrometric Analysis of Exhaled Breath for the Identification of Volatile Organic Compound Biomarkers in Esophageal and Gastric Adenocarcinoma. Annals of Surgery. 2015 Dec;262(6):981-990.

d. Kamal, F., Kumar, S., Edwards, M. R., Veselkov, K., Belluomo, I., Kebadze, T., Romano, A., Trujillo-Torralbo, M. B., Walton, R., Ritchie, A. I., Laponogov, I., Donaldson, G., Wedzicha, J. A., Johnston, S.L., Singanayagam, A., & Hanna, G. B. (2021). Virus-induced volatile organic compounds are

detectable in exhaled breath during pulmonary infection. American Journal of Respiratory and Critical Care Medicine, 204(9), 1075–1085.

3) BASIS: AI-Driven Platform for Chemically Augmented Histology and Mass Spectrometry Imaging. While digital pathology has transformed diagnostics, it remains largely limited to cellular morphology. I led the development of BASIS, an open-source, high-performance AI platform designed to bridge this gap by integrating Mass Spectrometry Imaging (MSI) into the pathology workflow. MSI provides in-depth molecular signatures that "augment" traditional histology with single cell proteomics and metabolomics data, but it generates massive datasets - often several hundred gigabytes per tissue section. The BASIS platform provides the first scalable architecture capable of processing thousands of MSI samples, automating the pipeline from raw data ingestion to molecular annotation and pattern discovery. By integrating omics data with morphological data and clinical metadata, BASIS transforms unrefined data into actionable clinical insights. My team has demonstrated the platform's utility in identifying lymph node metastases in gastric and esophageal cancers, performing 3D metabolic reconstructions, and quantifying intratumor heterogeneity.

Selected Publications:

a. Veselkov, K., * Sleeman, J., Claude, E., Vissers, J. P. C., Galea, D., Mroz, A., Laponogov, I., Towers, M., Tonge, R., Mirnezami, R., Tak.ts, Z., Nicholson, J. K., & Langridge, J. I. (2018). BASIS: High performance bioinformatics platform for processing large-scale mass spectrometry imaging data in chemically augmented histology. Scientific Reports, 8(1), 4053. *Corresponding author.

b. Inglese, P., McKenzie, J. S., Mroz, A., Kinross, J., Veselkov, K., Holmes, E., Tak.ts, Z., Nicholson, J.K., & Glen, R. C. (2017). Deep learning and 3D-DESI imaging reveal hidden metabolic heterogeneity of cancer. Chemical Science, 8(5), 3500–3511.

c. Abbassi-Ghadi, N., Golf, O., Kumar, S., Antonowicz, S., McKenzie, J. S., Huang, J., Strittmatter, N., Kudo, H., Jones, E. A., Veselkov, K., Goldin, R., Tak.ts, Z., & Hanna, G. B. (2016). Imaging of esophageal lymph node metastases by desorption electrospray ionization mass spectrometry. Cancer

Research, 76(19), 5647–5656.

d. Veselkov, K. A.*, Mirnezami, R., Strittmatter, N., Goldin, R. D., Kinross, J., Speller, A. V., Abramov, T., Jones, E. A., Darzi, A., Holmes, E., Nicholson, J. K., & Tak.ts, Z. (2014). Chemo-informatic strategy for imaging mass spectrometry-based hyperspectral profiling of lipid signatures in colorectal cancer. Proceedings of the National Academy of Sciences, 111(3), 1216–1221. *Corresponding author

4) HyperFoods/DREAMLAB: Crowdsourced Supercomputing for Food Molecular Discovery and the Food-Genome Interactome. Mapping the millions of interactions between food molecules, drugs, and the human genome requires immense computational power that is often prohibitively expensive or unavailablethrough traditional institutional clusters, slowing the discovery of preventative nutritional strategies. I pioneered the HyperFoods and DREAMLAB initiatives to bypass these hardware bottlenecks by architecting a "virtual supercomputer" that harnessed the idle processing power of 250,000+ smartphones globally. Using this crowdsourced infrastructure, my team deployed graph-based deep learning to interrogate multi-omics datasets and map the "food-genome interactome." This allowed us to identify thousands of "cancer-beating" molecules within common foods and repurpose existing drugs for indications in oncology and COVID-19. Recognized by BBC Click and Sky Swipe, this work provides a scalable framework for precision nutrition and represents one of the world's largest citizen-science applications in medicine.

Selected Publications:

a. Veselkov, K., * Gonzalez, G., Aljifri, S., Galea, D., Mirnezami, R., Youssef, J., Bronstein, M., & Laponogov, I. (2019). HyperFoods: Machine intelligent mapping of cancer-beating molecules in foods. Scientific Reports, 9(1), 9237. Corresponding author. Journal top 100 collection. *Corresponding author.

b. Laponogov, I., Gonzalez, G., Shepherd, M., Qureshi, A., Veselkov, D., Charkoftaki, G., Vasiliou, V., Youssef, J., Mirnezami, R., Bronstein, M., & Veselkov, K.* (2021). Network machine learning maps phytochemically rich “hyperfoods” to fight COVID-19. Human Genomics, 15(1), 1. *Corresponding author.

c. Southern, J., Gonzalez, G., Borgas, P., Poynter, L., Laponogov, I., Zhong, Y., Mirnezami, R., Veselkov, D., Bronstein, M., & Veselkov, K.* (2023). Genomic-driven nutritional interventions for radiotherapyresistant rectal cancer patient. Scientific Reports, 13(1), 14862. *Corresponding author

d. Rita, L., Neumann, N. R., Laponogov, I., Gonzalez, G., Veselkov, D., Pratic., D., Aalizadeh, R., Thomaidis, N. S., Thompson, D. C., Vasiliou, V., & Veselkov, K.* (2023). Alzheimer’s disease: Using gene/protein network machine learning for molecule discovery in olive oil. Human Genomics, 17(1), 57. *Corresponding author

5) CATALYST: Causally-Augmented Therapeutics and Graph Learning Systems. Biological systems are defined by complex, non-linear interactions where a single perturbation - whether a drug, a nutrient, or a genetic mutation - can have cascading effects across the human interactome. Traditional machine learning models often fail to capture these relational structures or distinguish between simple correlations and true causal drivers of disease, limiting our ability to predict the efficacy of combinatorial therapies or dietary interventions. I spearheaded the development of the CATALYST framework, utilizing Graph Neural Networks (GNNs) and causally inspired AI architectures to model the complex dependencies within biological networks. By integrating graph representation learning with causal inference, my team developed methods for the combinatorial prediction of therapeutic perturbations, allowing for the simulation of how multi-agent "cocktails" affect disease states. We applied these models to identify novel nutraceutical targets for Polycystic Ovary Syndrome (PCOS) and developed LLM-driven strategies for optimizing ingredient substitutions to maximize phytochemical density. This moves AI beyond classification toward "generative discovery," where models

suggest optimal interventions to steer molecular profiles toward health.

Selected Publicaitons:

a. Gonzalez, G., Lin, X., Herath, I., Veselkov, K., Bronstein, M., & Zitnik, M. (2025). Combinatorial prediction of therapeutic perturbations using causally inspired neural networks. Nature Biomedical Engineering. 2025; 1-18.

b. Hanassab, S., Southern, J., Olabode, A. V., Laponogov, I., Bronstein, M., Comninos, A. N., Heinis, T., Abbara, A., Izzi-Engbeaya, C., Veselkov, K*., & Dhillo, W. S. (2025). Identifying nutraceutical targets to treat polycystic ovary syndrome using graph representation learning. npj Women’s Health, 3(1), 68. *Corresponding author

c. Rita, L., Southern, J., Laponogov, I., Higgins, K., & Veselkov, K.* (2024). Optimizing ingredient substitution using large language models to enhance phytochemical content in recipes. Machine Learning and Knowledge Extraction, 6(4), 2738–2752. *Corresponding author

d. Gonzalez, G., Gong, S., Laponogov, I., Bronstein, M., & Veselkov, K. * (2021). Predicting anticancer hyperfoods with graph convolutional networks. Human Genomics, 15(1), 33 *Corresponding Author.