All of these efforts paid off in the end, but at the start, we didn’t know whether any of them would. That is the purpose of research, gunning for the impossible.
#IRIS #singlecell #microfluidics #teamwork
Preprint: www.biorxiv.org/content/10.1...
Timothée Ferrari | IRIS image pre-processing in the age of AI, that should be easy, right? It wasn’t. Tim manually annotated tens of thousands of cell images obtained with IRIS for correct focal planes to train #YOLO enabling automated image pre-processing (Supp. Fig. 3B).
Nadia Grenningloh | While deterministic barcoding worked well enough for HEK293T cells, performance for PBMCs was initially abysmal - fixed with hard work (Supp. Fig. 2C–J). The cost: more than 50 non-optimized IRIS runs, each involving >12-hour days.
Camille Lambert | Building the deterministic microfluidics required for precision stopping, cell encapsulation, and droplet sorting did not fall from the sky. It was an iterative process at CMi@EPFL. Iteration #42 finally did the job (Supp. Fig. 2A & 3A).
Getting integrated imaging and sequencing per cell off the ground with IRIS took a lot of grind. Countless iterations across every component. 📸+🧬=😍
I want to highlight three key craftspeople who built, established, and coded what made IRIS work (preprint at thread end).
Link to preprint: 🦾
www.biorxiv.org/content/10.1...
When IRIS was just a word on a whiteboard, they and many others started building from scratch with (extremely) limited budget. Their hard work was pivotal in getting IRIS off the ground.
#microfluidics #singlecell #phenomics #imaging
Assia Ouanaya automated library-prep steps, saving ~4h per experiment.
Maximilian Kohnen built IRIS’s first functional custom optics for focal-plane imaging.
@carolinewandinger.bsky.social developed the scRNA-seq mapping and QC pipeline that still runs smoothly every day.
Today, we use their work every day.
Romina Augustin established well-coding the biochemical foundation that enabled our ‘droplet consortia’.
Katharina Eckstein removed five PCR cycles from the workflow, effectively tripling gene-detection sensitivity.
Building IRIS meant turning a 'simple' idea into an integrated deterministic microfluidic instrument that combines imaging<>transcriptomes per cell. Those five years were full of “0→1” moments; each essential for making IRIS reality.
One long enjoyable haul! 🏋️♂️
This collaboration made our journey truly enjoyable (5 years of #TechDev + 2 years of biology).
IRIS was built in the lab of
@bartdeplancke.bsky.social
.
More stories on how IRIS came to be are coming up. 💘
@Wouter Karthaus pointed us to the cell-cycle space and the intricacies of the FUCCI reporter system > making the DREAM (complex) tangible (Figures 4,5).
@mariabrbic.bsky.social a Brbic brought new ML/AI angles to leverage images and transcriptomes per cell at scale.
Once IRIS was up and running: We wondered what biology should we focus on?
@Berend Snijder pushed us toward understanding the molecular forces shaping functionally distinct naïve T-cell architectures (Figure 6).
Now IRIS brings imaging<>transcriptome data together on a per-cell basis | #phenomic data at scale bridging the gap between molecular signatures and functional phenotype.
We spent five years building a technology rooted in who we are: Techies.
www.biorxiv.org/content/10.1...
