(Senior) ML Researcher – Molecular Privacy

At Apheris, we power federated data networks in life sciences to address the data bottleneck in training highly performant machine learning models. Publicly available molecular datasets are insufficient to train models that meet real industry requirements. Our product enables biopharma organizations to collaboratively train higher-quality models on their combined data, while ensuring that data ownership, IP, and governance remain with the
 original custodians. Our federated computing infrastructure is designed with privacy and control as first-class concerns.
As we double down on structural biology and ADMET as core areas within our drug discovery work, we are looking for a privacy-focused Senior ML Engineer to take technical ownership of privacy risk assessment & mitigations within our federated modelling initiatives. This is a hands-on, high-impact role centred on understanding how real drug discovery models behave in practice, identifying where privacy risks emerge, and generating empirical evidence to assess and mitigate those risks.
You will work within our AI Applications Engineering team and act as a technical authority on privacy for machine learning in drug discovery. A key part of the role is mapping real-world model usage to concrete threat models and experimental designs and clearly communicating the resulting evidence and conclusions to external partners, consortium stakeholders, and internal leadership.
You should bring strong hands-on experience with machine learning models used in drug discovery—particularly structure-based and protein–ligand modelling, with exposure to adjacent areas such as ADMET. You should be comfortable working directly with modelling pipelines, uncertainty estimation, and model outputs to reason about privacy risk, rather than treating privacy as a theoretical or policy-driven concern.
If you want to be part of a mission-driven team building federated AI systems for life sciences, and you are motivated by turning complex modelling behaviour into clear, defensible privacy conclusions for high-stakes collaborations, this role is for you.

• Design and execute practical privacy risk experiments on real drug discovery models, mapping theoretical threats to realistic attack surfaces.
• Work hands-on with molecular and structural ML pipelines (e.g. protein–ligand models, co-folding architectures, ADMET / QSAR data) to identify how modelling choices, representations, and uncertainty exploration can expose sensitive signal.
• Build and adapt experimental tooling for privacy analysis, including uncertainty probing, generative reconstruction tests, and distributional leakage experiments.
• Generate technically credible privacy evidence through hands-on modelling and experimentation, and convert that evidence into clear, informative reports and presentations for consortium and customer decision-makers.
• Translate empirical findings into clear, technically credible privacy narratives for customers, internal stakeholders, and partner organizations.
• Collaborate closely with ML engineers, scientific teams, and other privacy stakeholders to design mitigation strategies that are grounded in actual model behaviour and implementation constraints.

By month 3:

• Develop a working understanding of Apheris’ product, federated training setup, and key life-sciences modelling use cases.
• Reproduce and extend at least one existing modelling pipeline to establish a baseline privacy and attack-surface assessment.
• Contribute to privacy analysis for one or more active federated drug discovery programs as they transition from setup into live operation.

 By month 6:

• Design and run practical privacy experiments on live federated modelling workflows (e.g. co-folding, binding, screening, ADMET), focusing on realistic leakage and attack scenarios.
• Generate quantitative and qualitative evidence and synthesize it into clear reports and slide decks for external and internal stakeholders.
• Act as a technically credible privacy counterpart in active program discussions across scientific, engineering, and governance audiences.

 By month 12:

• Own privacy risk assessment and evidence generation across multiple federated drug discovery network cases as programs scale and evolve.
• Identify and evaluate privacy risks that emerge only under real operational conditions, and propose grounded mitigation strategies.
• Shape organizational standards for privacy experimentation, interpretation, and communication in long-running drug discovery networks.

You should apply if

• You have deep hands-on experience building and modifying machine learning models in drug discovery, particularly structure-based modelling and co-folding, with exposure to adjacent areas such as ADMET.
• You have hands-on experience with privacy for machine learning and/or federated learning, including reasoning about privacy risk, model behaviour, and governance in distributed or multi-party settings.
• You are comfortable designing empirical privacy experiments and drawing defensible conclusions from quantitative and qualitative evidence.
• You can communicate complex technical risks clearly and credibly to senior scientific, technical, and leadership stakeholders.
• You are comfortable owning ambiguous, cross-cutting problems end to end and setting direction as well as executing.

• You’ve published or led substantial technical work in machine learning or computational biology, with contributions in venues such as NeurIPS, ICML, ICLR, Nature Methods, Bioinformatics, or equivalent industry research outputs.
• You have operated in industry consortia or multi-organization collaborations, and understand the technical, political, and governance dynamics they impose.
• You have helped define, defend, or standardize privacy or risk positions in customer-, partner-, or regulator-facing contexts.
• You have acted as a technical authority, shaping standards, frameworks, or long-term direction rather than working solely against a predefined brief.

• Industry-competitive compensation, incl. early-stage virtual share options
• Remote-first working – work where you work best, whether from home or a co- working space near you
• Great suite of benefits, including a wellbeing budget, mental health benefits, a work- from-home budget, a co-working stipend and a learning and development budget
• Regular team lunches and social events
• Generous holiday allowance
• Quarterly All Hands meet-up at our Berlin HQ or a different European location
• A fun, diverse team of mission-driven individuals with a drive to see AI and ML used for good
• Plenty of room to grow personally and professionally and shape your own role

Our interview process is split into three phases:

• Initial Screening: If your application matches our requirements, we invite you to an initial video call to explore the fit. In this 30-minute interview, you will get to know us and the role. The interviewer will be interested in your relevant experiences and skills, as well as answer any question on the company and the role itself that you may have.
• Take-home exercise: You will be asked to undertake a short, offline coding assessment and prepare a case-study to be presented in the next stage.
• Presentation & Deep Dive: In this phase, we will ask you to give a short presentation on your case-study topic. Domain experts from our team will assess your skills and knowledge required for the role by asking you about meaningful experiences or your solutions for specific scenarios in line with the role we are staffing.
• Final Interview: Finally, we invite you for to meet with our founders, talking about our culture and meeting future co-workers on the ground.