Data scientist

Technical skills: image/signal analysis, Deep-learning engineering, software development

Experience

Post-doctoral researcher @ Centre Interdisciplinaire de Nanosciences de Marseille (CINaM) (June-July 2024)

• Development of appropriate benchmarks to test segmentation, classification and regression performance of traditional and Deep-learning models on test data
• Internship supervision to design cell pair descriptors and an intuitive viewer to interact with cell pairs

Ph.D. candidate @ Laboratoire Adhésion & Inflammation (LAI) and CINaM (September 2020 - April 2024)

• Data analysis: quantified and modeled the increase in immune-to-cancer-cell kill rate with new antibodies from multichannel optical microscopy movies
• Scientific discovery: described and measured the spreading decision rate of immune cells on antibody-covered surfaces that correlates with the kill rate
• Technical but accessible: assembled Celldetective, a versatile software developed organically with and for collaborators to perform the studies mentioned above autonomously. Go-to person, internally, to design image/signal/time-series analysis pipelines

Projects

You will find more numerical projects on the GitHub repository. Here are three highlights where I collaborated directly with experimental scientists, giving them keys to analyze their data.

Celldetective (Source code | Preprint)

Goal: help experimentalists with low to no coding skills measure cell interactions from microscopy images with single-cell resolution

Challenges: mixed cell populations, high density, heterogeneous and partial fluorescence marking, imaging and experimental conditions varying regularly, important data volume (~50-150 Gb per experiment)

Techniques:

• traditional or Deep-learning segmentation (StarDist, Cellpose), with dataset curation and model training
• cell tracking with a Bayesian tracker (bTrack)
• event class formulation to describe the dynamic cell states. Automation with convolutional models interpreting single-cell signals (classification, regression)
• neighborhood schemes to link the single-cells of a population (e.g. cancer cells) to the single-cells of the other population (e.g. immune cells) and describe cell pairs … and many more interesting techniques!

Delivery: a Python package accompanied by a complete graphical interface, forming a closed and intuitive ecosystem to correct images, segment, track & measure cells, pick up events, compute neighborhoods, plot results (survival functions, measurement distributions, collapse single-cell signals with respect to an event time)…

Developed in close collaboration with experimentalist researches (Ph.D. students, engineer, post-doc). Software routinely used in LAI & CINaM on several projects.

Yeast cell detection (Publication)

Goal: measure the enrichment of non-fluorescent yeast cells from optical microscopy images (brightfield and fluorescence) delayed temporally (projection is not enough)

Techniques:

1. traditional segmentation of the yeast cells from brightfield with top-hat filtering and thresholding
2. detection of fluorescent yeast cells with TrackPy (tracker)
3. linking of brightfield-detected yeast cells to fluorescent yeast cells with a co-distance matrix (closest neighbor within a critical distance)

Delivery: a Fiji macro for 1), a PyQt GUI to combine the measurements of 1) with 2) and perform task 3), all the way to plotting results

Traction force microscopy image registration (Source code | Preprint)

Goal: perform a subpixel registration of traction force microscopy images to be able to detect subtle bead motion due to cells pulling on gels

Techniques:

1. single-particle tracking data analysis to estimate the drift with respect to a reference frame (moving reference if needed)
2. apply an inverse shift in Fourier space to achieve a subpixel correction with no artefacts

Delivery: an easy-to-use Jupyter notebook and Python package

Original TFM stack	Drift corrected stack

Talks

Education

• Ph.D. Biophysics, Aix-Marseille Université, France (April 2024)

• M.S. Physics, Aix-Marseille Université, France (June 2020)

• B.S. Physics, University of Calgary, Canada / Aix-Marseille Université, France (May 2018)

Publications

Quantitative Microscopy for Cell–Surface and Cell–Cell Interactions in Immunology B. Diaz-Bello, D. El Arawi, R. Torro, P. Chames, K. Sengupta, L. Limozin bio-protocol 2025 [ BibTex | ArXiv ]
Transfer of polarity information via diffusion of Wnt ligands in C. elegans embryos P. Recouvreux, P. Pai, V. Dunsing, R. Torro, M. Ludanyi, P. Mélénec, M. Boughzala, V. Bertrand, PF. Lenne Current Biology 2024 [ BibTex | PDF | ArXiv ]
Celldetective: an AI-enhanced image analysis tool for unraveling dynamic cell interactions R. Torro, B. Diaz-Bello, D. El Arawi, L. Ammer, P. Chames, K. Sengupta, L. Limozin bioRxiv 2024 [ BibTex | ArXiv ]
Antigen density and applied force control enrichment of nanobody-expressing yeast cells in microfluidics M. Sanicas, R. Torro, L. Limozin, P. Chames Lab on a Chip 2024 [ BibTex | PDF | ArXiv ]
Cellular forces during early spreading of T lymphocytes on ultra-soft substrates F. Mustapha, M. Biarnes-Pelicot, R. Torro, K. Sengupta, PH. Puech bioRxiv 2022 [ BibTex | PDF | ArXiv ]
Radical-assisted polymerization in interstellar ice analogues: formyl radical and polyoxymethylene T. Butscher, F. Duvernay, G. Danger, R. Torro, G. Lucas, Y. Carissan, D. Hagebaum-Reignier, T. Chiavassa Monthly Notices of the Royal Astronomical Society 2019 [ BibTex ]