Technology: Small Data AI
Our technology, called Small Data AI unifies 3 key capabilities: (1) structural representations of biomolecules, (2) self/semi-supervised learning from partially labelled data with less than 10k labels, and (3) motif detections in structural representations. Feel free to click on each step for further information.
Using structural representations of data enables learning in small data scenarios as structure diversity alone carries information that can be harnessed to make predictive models. Additionally, structural models can be queried for “explainability” so that our models are not black box.
Mechanochemical Crosstalk Produces Cell-Intrinsic Patterning ...
Here we built a multimodal mechanochemical model to recapitulate a complex experimental workflow, leading to the discovery of an intimate relationship between biochemical signalling and the mechanics of cell division in cancer
Dimitracopoulos et al., Curr Biol, 2020.
FeII4L4 Tetrahedron Binds to Nonpaired DNA Bases
A core hypothesis behind our approach is to think about oligonucleotide therapeutics in terms of thermodynamics. In this paper we designed and characterised oligonucleotides and leveraged this knowledge to build machine learning algorithms.
Zhu et al., JACS, July 2019.
KymoButler, a deep learning software for automated kymograph analysis
Our seminal paper on the analysis of line-like structures in images. The original model was trained on just 100 images and is now used by hundreds of researchers around the world.
Jakobs et al., eLife, August 2019.
Receptor-specific interactome as a hub for rapid cue-induced selective translation in axons
Koppers et al., eLife, November 2019.
Single-molecule analysis of endogenous β-actin mRNA trafficking reveals a mechanism for compartmentalized ...
Turner-Bridger et al., PNAS, September 2018.
Cortical Cell Stiffness Is Independent of Substrate Mechanics
Rheinlaender et al., Nature Material, 2020
Mitotic Rounding Alters Cell Geometry to Ensure Efficient Bipolar Spindle Formation
Lancaster et al., Dev Cell, April 2013.
Late Endosomes Act as mRNA Translation Platforms and Sustain Mitochondria in Axons
Cioni et al., Cell, January 2019.
Select publications by our team
Explore biochemical space with
Our platform uses semi supervised learning to predict unknown data endpoints in hypothetical biochemical space
Search for sub-structures that
correlate with endpoints
Our platform can find structural features that explain endpoints e.g. regions in images or motifs and substructures in molecules
Predict molecular graphs for natural representation
Data with unknown
Data with known
Our platform first builds graph representations of molecule and image data for both data with known (blue) and unknown (light blue) endpoints.