Job description

ABOUT ABSENTIA LABS Absentia Labs is building mechanistic AI models to predict how drug compounds will behave in the human body before costly preclinical and clinical studies. Our platform integrates molecular properties, exposure, biological context, experimental evidence, and machine learning to identify potential safety liabilities earlier and explain the biological mechanisms driving them. We initially focus on predictive toxicology and drug safety, with the broader goal of building foundational models that can reason across human biology, pharmacology, and translational outcomes. We work at the intersection of frontier AI, computational biology, chemistry, toxicology, and drug development. THE ROLE We are looking for an AI Research Scientist to help advance the modeling approaches at the core of Absentia's platform. This is a research role for someone who wants to develop new machine learning methods for difficult scientific problems, not simply apply existing models to biological datasets. You will formulate research questions, design and run experiments, develop novel model architectures and learning strategies, and investigate how models can integrate heterogeneous biological and chemical evidence to predict complex human outcomes. Our research problems span molecular representation, graph learning, transformers, multimodal learning, representation learning, uncertainty, mechanistic reasoning, and biological generalization. You will work closely with our CTO, AI/ML engineers, data engineers, and scientists to move promising ideas from research hypotheses into validated modeling capabilities. The goal is straightforward but difficult: develop AI systems that can reason about how a drug interacts with human biology well enough to make useful predictions before those outcomes are observed experimentally or clinically. WHAT YOU’LL DO - Own ambitious research problems in predictive toxicology, drug safety, and computational biology from hypothesis through experimental validation. - Develop and evaluate novel deep learning architectures and training methods, including graph neural networks, transformers, multimodal models, generative approaches, and other emerging architectures where scientifically appropriate. - Investigate representations that connect molecular structure, biological targets, pathways, dose and exposure, pharmacology, toxicology, and clinical outcomes. - Develop approaches for learning from heterogeneous, sparse, noisy, and partially observed scientific datasets. - Explore methods for cross-domain and out-of-distribution generalization, including prediction on novel compounds, chemical spaces, biological contexts, and endpoints. - Design rigorous experiments, benchmarks, ablations, and evaluation frameworks that distinguish genuine biological generalization from memorization or dataset artifacts. - Develop methods for uncertainty estimation, calibration, applicability-domain assessment, and confidence-aware prediction. - Investigate approaches for making model predictions more mechanistically interpretable, including identifying biological pathways, targets, systems, and evidence contributing to predicted outcomes. - Explore multimodal and foundation-model approaches capable of combining chemical, biological, experimental, literature-derived, and clinical evidence. - Identify limitations and failure modes in existing models and turn those observations into new research directions. - Work with AI/ML and data engineers to translate successful research into reproducible, scalable modeling systems. - Contribute to Absentia's scientific strategy, including validation studies, external scientific collaborations, publications, and research supporting regulatory evaluation of our models. RESEARCH PROBLEMS YOU MIGHT WORK ON Rather than hiring against a single architecture, we're interested in researchers who can attack questions such as: 1. How should a model represent a drug? 2. Can molecular graphs, learned embeddings, biological targets, metabolites, and pharmacological context be represented jointly rather than as independent features? 3. How do we model exposure? 4. Can models reason about how dose, route of administration, metabolism, tissue exposure, and PK/PD change the probability and mechanism of toxicity? 5. How do we predict beyond the training distribution? 6. How can we determine whether a prediction for a novel compound represents genuine biological generalization rather than interpolation over known chemistry? 7. How do we connect mechanisms to outcomes? 8. Can models learn relationships between molecular interactions, pathways, organ systems, adverse events, and clinical outcomes? 9. How should uncertainty propagate through biological predictions? 10. Can we distinguish uncertainty caused by limited chemical similarity, uncertain biological evidence, exposure assumptions, or endpoint ambiguity? 11. Can one model reason across organ systems? 12. How should liver, cardiac, renal, and other biological systems eventually interact within a broader model of human drug response? WHO YOU ARE You are a researcher who is comfortable working where the correct modeling approach is not yet known. You care about understanding why a model works, where it fails, and whether it is actually learning something generalizable. You are comfortable moving between mathematical ideas, experimental code, scientific literature, and large-scale empirical results. You can pursue a research direction independently, but you also enjoy working closely with engineers and domain scientists to turn research into systems that matter. Most importantly, you want your research to have consequences beyond a benchmark. YOU LIKELY HAVE - A PhD in machine learning, artificial intelligence, computer science, computational biology, computational chemistry, applied mathematics, statistics, or a closely related field, or equivalent demonstrated research experience. - A strong research track record demonstrated through publications, novel methods, significant open-source research, or technically substantial research projects. - Deep understanding of modern machine learning and deep learning. - Hands-on experience developing and evaluating models in PyTorch, JAX, or equivalent frameworks. - Experience with one or more of: graph neural networks, transformers, representation learning, multimodal learning, generative modeling, self-supervised learning, probabilistic modeling, or foundation models. - Strong experimental instincts and experience designing controlled evaluations, ablations, and reproducible research. - Ability to reason carefully about dataset construction, leakage, confounding, generalization, and evaluation methodology. - Ability to independently identify important research questions and carry projects from initial hypothesis through rigorous experimental results. - Strong written and verbal communication skills. BONUS IF YOU HAVE You do not need to come from drug development or toxicology. We care more about exceptional research ability and the capacity to learn new scientific domains. - Computational biology, computational chemistry, cheminformatics, or drug discovery. - Molecular machine learning or geometric deep learning. - Biological foundation models. - Multimodal scientific machine learning. - Causal inference or mechanistic modeling. - Uncertainty quantification and model calibration. - Out-of-distribution/generalization research. - Active learning or experimental design. - Pharmacology, toxicology, PK/PD, or translational science. - Working with large scientific or biomedical datasets. - Research spanning machine learning and the natural sciences. WHAT WE OFFER - Competitive compensation, including meaningful equity participation, allows you to share directly in the long-term success and growth of the company. - The opportunity to work on foundation-level ML systems applied to real scientific problems. - Ownership over model design and training strategy, not just implementation. - Close collaboration with data, AI/ML, infrastructure, and scientific teams. - High autonomy, low bureaucracy, and a culture that values technical depth. - Flexible remote or hybrid work arrangements. HOW TO APPLY Please submit your resume and a brief note describing your experience training large-scale models. Links to GitHub repositories, papers, or technical write-ups are encouraged. OUR COMMITMENT Absentia Labs is an equal opportunity employer. We believe diverse teams build better systems and stronger science, and we encourage applicants from all backgrounds to apply.

Requirements

Must-haves

We are looking for an AI Research Scientist to help advance the modeling approaches at the core of Absentia's platform.
This is a research role for someone who wants to develop new machine learning methods for difficult scientific problems, not simply apply existing models to biological datasets.
Our research problems span molecular representation, graph learning, transformers, multimodal learning, representation learning, uncertainty, mechanistic reasoning, and biological generalization.
You will work closely with our CTO, AI/ML engineers, data engineers, and scientists to move promising ideas from research hypotheses into validated modeling capabilities.
Own ambitious research problems in predictive toxicology, drug safety, and computational biology from hypothesis through experimental validation.
Investigate representations that connect molecular structure, biological targets, pathways, dose and exposure, pharmacology, toxicology, and clinical outcomes.
Develop approaches for learning from heterogeneous, sparse, noisy, and partially observed scientific datasets.
Explore methods for cross-domain and out-of-distribution generalization, including prediction on novel compounds, chemical spaces, biological contexts, and endpoints.
Design rigorous experiments, benchmarks, ablations, and evaluation frameworks that distinguish genuine biological generalization from memorization or dataset artifacts.
Develop methods for uncertainty estimation, calibration, applicability-domain assessment, and confidence-aware prediction.
Investigate approaches for making model predictions more mechanistically interpretable, including identifying biological pathways, targets, systems, and evidence contributing to predicted outcomes.
Explore multimodal and foundation-model approaches capable of combining chemical, biological, experimental, literature-derived, and clinical evidence.
Identify limitations and failure modes in existing models and turn those observations into new research directions.
Work with AI/ML and data engineers to translate successful research into reproducible, scalable modeling systems.
Contribute to Absentia's scientific strategy, including validation studies, external scientific collaborations, publications, and research supporting regulatory evaluation of our models.
Rather than hiring against a single architecture, we're interested in researchers who can attack questions such as:
How should a model represent a drug?
Can molecular graphs, learned embeddings, biological targets, metabolites, and pharmacological context be represented jointly rather than as independent features?
How do we model exposure?
Can models reason about how dose, route of administration, metabolism, tissue exposure, and PK/PD change the probability and mechanism of toxicity?
How do we predict beyond the training distribution?
How can we determine whether a prediction for a novel compound represents genuine biological generalization rather than interpolation over known chemistry?
How do we connect mechanisms to outcomes?
Can models learn relationships between molecular interactions, pathways, organ systems, adverse events, and clinical outcomes?

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aimachine learningml

Platform: ashby
Effort: quick