Data Scientist

About the Role

We are an interdisciplinary research team at the Cancer Research Center, Sheba Medical Center, working at the intersection of machine learning and genomic medicine. Our work applies large language models to large-scale biological data, with the goal of advancing precision medicine. The team is multidisciplinary and internationally connected, with access to advanced compute resources and unique large-scale datasets. (Further details about the research are shared with candidates during the process.)

●    Design, pre-train, and fine-tune large-scale language models for biological and genomic data.

●    Research and develop novel representation and tokenization strategies.

●    Experiment with state-of-the-art architectures, including long-context modeling.

●    Build rigorous evaluation and benchmarking pipelines.

●    Run distributed training on multi-GPU environments , optimizing performance, memory, and throughput.

●    Collaborate closely with biologists, clinicians, and researchers.

●    Contribute to scientific publications at leading venues.

●    MSc or PhD in Computer Science, Mathematics, Statistics, Engineering, Physics, Bioinformatics, or a related quantitative field — or equivalent proven experience.

●    Demonstrated hands-on experience developing and training deep learning models, ideally LLMs or Transformer-based networks.

●    Strong proficiency in Python and a major deep learning framework (PyTorch and/or JAX).

●    Deep theoretical understanding of:

○    The Transformer architecture and attention mechanisms (including variants such as FlashAttention, RoPE/ALiBi).

○    Self-supervised training methods — masked language modeling, causal/next-token prediction, contrastive and embedding learning.

○    Optimization and training dynamics (learning-rate schedules, regularization, scaling laws, handling training instabilities).

○    Generalization and evaluation principles — overfitting, data leakage, metric selection (AUROC, AUPRC, MCC), and rigorous benchmarking.

●    Experience working in Linux environments and version control with Git.

●    Ability to work with large-scale data and design efficient data pipelines.

●    High proficiency in English (reading, writing, and scientific communication).

●    Familiarity with existing genomic / biological language models: DNABERT/DNABERT-2, Nucleotide Transformer, HyenaDNA, Caduceus, Evo, GENA-LM, and similar.

●    Experience with architectures for very long sequences: state-space models (Mamba), Hyena operators, and sub-quadratic attention methods.

●    Experience with distributed training and dedicated frameworks: FSDP / DeepSpeed, Megatron-LM, NVIDIA NeMo, as well as CUDA and mixed precision (bf16/fp8).

●    Background in biology, genetics, or genomics — understanding of variants, zygosity, sequencing, and omics analysis (genomics, transcriptomics, epigenomics). This is not a hard requirement — we are happy to teach the biological side to a strong algorithmic candidate.

●    Experience in variant effect prediction or representation learning for biological sequences.

●    Familiarity with frontier architectures such as JEPA and hierarchical models.

●    Publications at leading venues (NeurIPS, ICML, ICLR, or top bioinformatics journals).

●    The chance to work on a groundbreaking scientific problem with real clinical impact, at the global frontier of the field.

●    Access to advanced compute resources and unique large-scale datasets.

●    An interdisciplinary, internationally connected team and a rich research environment.

●    Research freedom, opportunities to publish and present at conferences, and professional growth.