Data Science & AI Innovation Postdoctoral Fellow Foundational & Multimodal Proteochemometrics models

Das ist der Job

Industrial large scale screening technologies such as DNA encoded libraries generate larger volumes of dense data.

Darum lohnt es sich

In this project we aim at training a large- proteochemometrics model based on high-throughput experimental data resulting from different read-out modalities and will assess how such models can be predictive for low-volume high quality affinity data with and without fine-tuning on a small number of such quality data points. /p pbWhy Join the Program? /b /p pThe Novartis Biomedical Research Postdoctoral Fellowship Program is designed to develop the next generation of scientific leaders, powering the future of medicine, through rigorous research, and immersive learning experiences, such as implementation of AI tools in biomedical research. /p pPostdoctoral Research Fellows benefit from: /p ul liGuidance from accomplished scientific leaders and subject matter experts /li liAccess to advanced technologies, platforms, and research capabilities /li liCollaboration across disciplines and organizational boundaries /li liA global and diverse community of postdoctoral fellows /li liDedicated programming designed to help fellows thrive throughout their careers. /li liPersonalized experiential learning opportunities through a Postdoc Practicum that empower fellows to explore new scientific domains, build cross-functional expertise, and expand their impact beyond their primary research project. /li liOpportunities to present research, publish in leading journals, and build an international scientific network /li /ul pWe are entering a new era of biomedical research breakthroughs through the convergence of biology, technology, and artificial intelligence tools, and fellows are also supported in engaging with these emerging approaches. /p pThis is a 100% training position of up to three years in duration. /p pbReimagining Medicine Together /b /p pAt Novartis, our purpose is to reimagine medicine to improve and extend people’s lives.

Through this program, you will grow as a scientist and future leader while contributing to discoveries that may ultimately benefit patients worldwide. /p pbKey Responsibilities /b /p ul liTrain and optimize foundational structure affinity models on high volume experimental data such as DNA encoded library screening data. /li liFine-tune models on high-quality low volume affinity data. /li liEvaluate the performance of the models with and without fine-tuning and compare it with the state-of-the-art models. /li liApply and evaluate promising model architectures prospectively in small molecule hit-finding projects. /li liEvaluate the performance of the models to predict off-targets for pharmacology safety assessments. /li /ul pbEssential Requirements /b /p ul liPhD (or equivalent doctoral degree) in a relevant scientific discipline completed prior to the fellowship start date.

The program is intended for scientists immediately following their PhD training (graduated in 2026) /li liDemonstrated record of scientific achievement (publications, presentations, patents, or equivalent) /li liStrong commitment to learning, innovation, and professional development Hands-on experiencewith cheminformatic workflows, and familiarity with descriptors and machine learning and deep learning in the context of cheminformatics. /li liExpertiseworking inLinuxhigh performancecomputing and cloud environments. /li liExpertisein Python scientific and deep learning stacks,familiarity with best practices in computational reproducible research (version control, testing, documentation). /li liExperience in training foundational models and / or processing huge datasets /li liDemonstrated abilityto work as part of an interdisciplinary team (i.e., biologists, chemists, data scientists), withproactive and results-orientedcommunication skills.Dedication to promotingmutualrespect,empathy,andpositivityin diverseprofessionalsettings. /li /ul pbDesirable Requirements /b /p ul liExperience with some of the following: ligand protein docking, ligand proteinco-folding, drug-target interaction models /li liExperience using synthons and transformations to generate virtual spaces, or to interrogate virtual spaces /li /ul /p #J-18808-Ljbffr ph3Summary /h3 pThe Novartis Biomedical Research Postdoctoral Fellowship Program offers a 3-year 100% position starting October 1, 2026, in Basel, Switzerland, focusing on foundational multimodal proteochemometrics models in drug discovery.

This opportunity enables early-career scientists to work with cutting-edge AI and biomedical research technologies in a collaboration with data and wet-lab scientists.

The fellows will train and optimize structure affinity models using high-throughput experimental data and evaluate their predictive performance for drug-target interactions in hit finding and safety assessments. /p h3About the Role /h3 pbAbout the Role /b /p pWe are excited to invite applications for the Novartis Biomedical Research Postdoctoral Fellowship Program, a unique training opportunity designed for exceptional early-career scientists eager to tackle some of the most challenging problems in biomedical research and drug discovery. /p pAs a Postdoctoral Research Fellow, you will join Discovery Sciences in Basel and pursue an innovative research project at the forefront of biomedical science and drug discovery.

You will work alongside leading scientists in a highly collaborative, multidisciplinary environment while gaining exposure to the broader ecosystem that translates scientific discovery into medicines. /p pOur fellows are empowered to ask bold scientific questions, apply cutting-edge technologies, and develop approaches that have the potential to transform patient care. /p pbResearch Opportunity /b /p pFoundational structure activity models that can predict small molecule protein interactions from the small molecule structure and protein sequence have attracted scientific attention.

However, the practical suitability of such models trained on public structure activity data has been limited as this data is both low in volume and extremely sparse.