Our focus is on progress of communication, discussion and sharing of ideas.
Many people's work has inspired our own, especially the prediction of gene expression on Spatial Transcriptomics technologies (e.g. see stNet - He et al 2020, and, HE2RNA - Schmauch et al 2020 ).
The idea of aligning cells to spatial histology slides is similar to Tangram (Biancalani et al 2020).
The inspiration for visualising single cell gene expression vectors has been borrowed from many pieces in the literature, most similar to the concept on 'basis vectors' for 'cellular identity' (Wagner et al 2016).
What it does
The model enables the transfer of information from spatial transcriptomics technologies (ST-tech) to single cell RNA-seq (scRNAseq) - essentially building a bridge based on transfer learning.
This allows for a two way approach that enables:
1 - Spatial to Gene expression: Alignment of a reference cell atlas to non-ST-tech histology slides, predicting gene expression
2 - Gene expression to Spatial: Transformation of gene expression only vectors to spatial image tiles
More information on the model and step-by-step process can be found at: link
How we built it
The process involved several stages, built on top of each other to provide the framework that we call module learning (modL).
Each stage required to be built and integrated into the next, learning parameters and passing them on to learn more parameters in a modular fashion - thus modular learning.
We use a Django framework hosted on a Ubuntu server, serviced by Google Cloud Platforms to build a Dash and Plotly dashboard.
Challenges we ran into
There were many challenges - combining and presenting models together in a modular fashion required the right tools and approaches.
By believing in consistent communication, discussion and sharing of ideas, the team managed to progress through the technical software and statistical problems.
Accomplishments that we're proud of
The importance of ST-tech has been highlighted in many regions of the scientific community.
We are especially proud of providing a framework that enables both researchers and medical practitioners to re-use ST-tech in different domains by sharing information from different sources.
What we learned
We learned to be persistent in our belief of our ideas, and to implement them technically in a way that allows for democratising technology - in a modular learning fashion that allows the community to update, adjust and improve each part of modL.
What's next for Reimagining resolution at single cells
We hope more collaborations can come from this work with our group at Explain.
There is intention to build upon this technology in an open-source environment.
We can be contacted at email@example.com. Please reach out to us for any interests.