Accelerating biological discoveries by machine learning and quantitative single particle microscopy

Advanced microscopic techniques produce vast amounts of unstructured data the analysis of which by conventional methodologies is tedious, time consuming and may be biased by unconscious biases. We have been developing agnostic quantitative and automated analysis methodologies based on machine learning to treat classify and annotate biological behaviors. The toolboxes offer rapid analysis often accelerated by 3-6 orders of magnitude, help eliminating potential human biases and provide statistical insights on biological parameters that underlie and control protein function and cellular responses.

High throughput single-nanocontainer readouts and deep learning:
Pushing the biomolecular recognition detection to new frontiers

Screening of biomolecular recognition often suffers from challenges such as long running time, high person power as well as excessive materials cost. To surpass these challenges have developed miniaturized assays for ultra-sensitive and high-throughput screening of biomolecular interactions and to explore:

a) DNA-DNA recognition and sub-attolitter cargo delivery,
b) Transporter function and
c) How membrane properties affect protein function.

Rapid and reliably analyis and classification of the multidimensional multi terabyte data is achieved by our deep learning analytic tools.

Metabolic pathways redefined:
Biased P450 metabolism by smart ligands targeting protein dynamics and targeting metabolic diseases

POR is a central molecular hub activating a plethora of metabolic pathways by donating electrons to more than 50 different cytochrome P450 enzymes (CYPs). Point mutations in POR cause severe metabolic disorder due to altered POR-CYP interactions. In collaboration with Paediatric Endocrinology at the University Hospital Bern, Switzerland, and Department of Plant Biology, University of Copenhagen, we study these interactions all the way from clinical phenotype down to the fundamental limit of individual proteins. Combining single molecule FRET and single turnover studies with cell studies and docking simulations we advance our understanding on the intricate role of conformational dynamics to activity and specificity and eventually how pathogenic mutations and small molecule ligand interactions control metabolic disorder and biosynthetic pathways.

Deciphering cellular choreography:
Insights into single particle of proteins, viruses, and pharmaceutics nanocarriers

We have pioneered the development of powerful methodologies to track the spatiotemporal localization in live cells of individual biomolecules,(proteins organelles viruses and nanocarriers) and quantify their interaction with membranes cell entry pathways and utilized this information to tailor their targeted delivery directly. To analyse the complex, multidimensional, multiterabyte data we acquire, we have employed novel methodologies based on machine learning that offer rapid precise and automated transition from raw microcopy images to quantitative biomedicine insights accelerating discoveries often by 104 times.

Bridging structure and function of CRISPR-Cas12a with smFRET and Cryo-EM"

Adaptive immunity in bacteria is accomplished by the CRISPR system, and CRISPR-associated proteins (Cas). Proteins coupled with RNA are guided by this system to recognize and cleave foreign genetic material. As such, it’s also a powerful method for genome editing, and is receiving lot of bio technical and medical attention currently. By using single molecule FRET we can study this system in great detail, and obtain a wealth of structural and kinetic information, when combining with other techniques. Read how we did this in Stella et al. (2018), published in Cell.