efroni labThe Systems Biomedicine Lab

what we work on

Research

Each of us carries millions of distinct T cell receptors. Together they record what the immune system has met and how it responded. We sequence that record, read tumor transcriptomes beside it, and build computational methods that turn these data into biology and, we hope, into better medicine.

The T cell repertoire

A T cell receptor is assembled by V(D)J recombination, and its CDR3 loop decides much of what the cell will recognize. We profile CDR3 sequences across thousands to millions of cells, produce repertoires from patients and model systems, and build tools to read them.

Blood tests for cancer

If a tumor leaves a mark on the T cells that circulate in blood, a blood sample should carry it. We train classifiers over peripheral repertoires and detect breast and ovarian cancer this way. In colorectal cancer we follow MAIT cells and patient risk.

The language of T cell receptors

We read CDR3 sequences as a language. Self-supervised models trained on large receptor collections expose the properties that decide T cell membership, and public receptors, shared between people, give a way into personal repertoires.

Repertoires in treatment and over time

Repertoires change with therapy and with time. A repertoire timestamp sits at the core of the response to CTLA-4 blockade, and after bone marrow transplantation CDR3 and V genes reconstitute along different paths.

Measuring repertoires well

Methods decide results. We compare repertoires extracted from RNA-seq with dedicated TCR sequencing, release tools and data, and take part in community benchmarks. Our solution won the AIRR-ML 2025 challenge.

Genomic networks

Genes act through pathways and networks. We combine pathway and network knowledge with gene and protein data to expose hidden properties of cell regulation. In cancer patient samples these methods point to the pathways most critical for diagnosis and prognosis, which we then study in the wet lab.

Pathways as biomarkers

We score whole pathways and networks instead of single genes. Pathway-level biomarkers predict, and help change, response to cancer therapy.

Variants and microRNAs that change drug response

A single nucleotide variant of PARP1 determines response to PARP inhibitors, and a BCL2 variant is associated with paclitaxel resistance across tumor types. In glioblastoma, miR-9 and the p38 network control cell mobility.

Single-cell transcriptomics

With Dr. Assaf Rotem and Dave Weitz's lab we set up a droplet microfluidics system that captures whole-transcriptome profiles of single cells. We use it to study epithelial cells and T cells in the tumor microenvironment, and we build methods for single-cell and bulk data.

Cell states from pathways

Pathway scores computed per cell place T cells in their states, and gene-to-gene coordination among single cells is lost with age.

Deconvolving bulk data

When only bulk RNA-seq exists, reference-free deconvolution estimates which cell types it came from. We also took part in a community assessment of deconvolution methods.

Where it began: executable models

The lab grew out of executable biology. Statechart models of thymocyte development, run and animated as reactive systems, showed how lineage decisions emerge from the behavior of many cells. That view of the immune system as a computing system still shapes how we read repertoires.

Thymocytes as reactive systems

Modeling, execution and visualization of T cell maturation in the thymus, and the emergent dynamics of lineage determination.

The immune system computes

Crowd wisdom, machine learning and immune cell reference repertoires as a way to think about how the body reads its own state.