As we interact with more and more Life Science Researchers using our tools, we are compiling a list of the most frequent (and hopefully most helpful) questions that we answer.
What is DeepMirror?
DeepMirror is the company that redefines your Life Science data analysis workflows with state-of-the-art Small Data Artificial Intelligence (AI) tools. With these tools you can build AI models from small Life Sciences datasets (10-100 datapoints), e.g. microscopy images, genetic sequences, amino acid sequences, small molecules etc. This means that one does not need access to large datasets and can just get going on small internal data. These models then live on our cloud and that you can be used used to run data analysis tasks (e.g. image analysis), find optimal experimental solutions (e.g., do an in silico RNA screen), and make discoveries (discover what substructure of an antibody is important for binding).
What is Small Data AI?
Most applications of AI to date were Big Data applications. That is applications, for which large quantities of labelled/validated data could be easily obtained. Consumer behaviour for examples generates massive datasets that can be easily used for AI. Furthermore, images required for autonomous driving tasks can be easily annotated by anyone. Beyond these Big Data applications however are millions of niche applications for which AI simply doesn’t work because validated data is impossible to come by. Think for example about defect detection in a specialist car manufacturing pipeline. Defects are super rare (1:1,000,000) and pipelines highly divergent between companies so that one often only has <100 validated defects to teach AI how to detect them.
In the Life Sciences this is especially problematic: Companies all have their own internal processes and collecting validated/labeled data requires week long experiments or paying experts high sums to manually go through the data. This means that conventional AI is simply not applicable to most bio datasets. We are here to change this with Small Data AI: A technology that uses semi- & self-supervised learning to learn from small validated datasets while also accessing raw unvalidated data (e.g. untested experimental settings or unlabelled images).
What can I use the platform for?
DeepMirror’s proprietary Small Data AI technology leverages small validated biomedical datasets together with large unvalidated datasets to enable Life Sciences teams to make accurate predictions, find optimal solutions, and make breakthrough discoveries.
Using as few as 100 validated datapoints, DeepMirror’s platform can:
- classify and detect objects/areas in images of biological tissues/cells
- predict the properties (e.g. activity & toxicity) of non-validated small-molecules
- predict antibody properties (e.g. binding) of non-validated peptides
- predict the gene-editing efficiency of non-validated gRNA sequences (CRISPR/Cas9)
- predict silencing efficacy of RNAi
- predict binding efficacy of primers
- accelerate high content screening of small molecules, genetic sequences, and proteins
DeepMirror is now open to pilot studies to solve Small Data problems in the Life Sciences with a focus on R&D and manufacturing.
What does the service entail?
We license access in form of a subscription model that gives you access to our cloud to build models, run inferences, and discover optimal solutions.
Can we publish our research if we used DeepMirror software for our analysis?
At DeepMirror we have an academic background, and we want the methods used in our software to be clearly explained in the publications that used it. Currently, we keep our AI training algorithms closed source but are very open with the models we use. If this is not possible (for example if the algorithm uses code from a differed closed source application) we will at least provide a description of how the software works in detail for the methods section of your publication.