AI/ML Revolutionizing Drug Discovery (2nd part)
AI/ML Revolutionizing Drug Discovery (2nd part)
Innovative TechBio Startups and Investors
For the first part of this newsletter:
"For a successful technology, reality must take precedence over public relations, for nature cannot be fooled."
Richard P. Feynman
⚙️ Artis Ventures
Artis Ventures founded in 2001 in California 🌵 is a venture capital firm in the emerging TechBio sector funding companies at the intersection of computer science and life science applying engineering principles and data-enabled discovery (Clinical Medicine, Synthetic & Computational Biology, Internet of TechBio and Data & Compute) to the healthcare space (Aging, Cardiopulmonary/ Metabolic, Consumer Health, Dermatology, Digestive Health/ Gastrointestinal, Immunology, Infectious, Neurology/ Mental Health, Oncology, Wellness and Women's Health).
Artis’ team has received notable industry recognition with inclusion in the Forbes Midas List, New York Times and CB Insights Top 100 Venture Capitalists, TechCrunch’s VC of the Year Award, SF Business Times 40 Under 40, and more. Stuart Peterson is the Founder and Managing Partner of ARTIS Ventures
Notable companies the firm has backed include: Stemcentrx, Lemonaid Health, Viacyte, Eko Health, Outpace Bio, RadAI, Delix Therapeutics, Freenome, Excision Bio, Modern Meadow, Palantir, YouTube, Cohesity and many more.
Artis Invests into ➡️
Fabric Genomics: a computational genomics company founded in 2019 and partnered in 2013 that provides end-to-end genomic data analysis, annotation, curation, classification and reporting solutions to clinical/hospital labs and life science companies. Fabric's analytic capabilities begin with raw data analysis and include the delivery of comprehensive insights from high throughput panels. On December 07, 2023, PlumCare RWE and Fabric Genomics partnered to deliver the Fabric AI platform with the PlumCare RWE FirstSteps newborn genome screening program in Greece.
RadAI: is empowering radiologists with AI to save time, reduce burnout and improve patient care. Founded in 2018 and partnered in 2021, the company so far developed Rad AI Reporting, Rad AI Omni Impressions, Rad AI Nexus and Rad AI Continuity. On January 16, 2024, Rad AI announced a partnership with Google Cloud in scaling its efforts to improve clinical radiologists’ workflows.
Cohesity: is an AI-powered data security and management organization, founded in 2013 and partnered in 2015.
ReplyBio: is a reprogramming biology company writing and delivering big DNA, incubated in 2020,
and many more.
On January 12, 2024, Artis Ventures announced its new class of AV Fellows selected from more than 400 applications, including 14 leading PhD, MD and postdoc candidates from world-renowned institutions globally, with a diverse range of backgrounds, including biophysics, cardiology, ophthalmology, bioengineering, cancer biology and more. The class will work closely with the ARTIS Ventures team to support all aspects of the investment process, from due diligence and deal sourcing to portfolio support.
Patrick Schwab is the Senior Director 🎩, Artificial Intelligence and Machine Learning at GSK
Pharmaceutical giant GSK is investing heavily in harnessing data and artificial intelligence for early-stage drug discovery, with a goal of hiring about 100 people this year to work on the effort, according to a company representative.
GSK, bolstered by the recent ascension of Chat GPT and natural language models, has brought on Danielle Belgrave as vice president of AI and ML. She was most recently at Google DeepMind, creating the health division with the DL unit.
⚙️ Pipe Bio
PipeBio founded in 2020 in Denmark 🇩🇰 is a cloud-based bioinformatics platform for biologics discovery, allowing wet lab scientists to easily analyze antibody and peptide sequences with functional assay data and bioinformaticians to deploy their own code and run workflows. They offer Antibody Discovery and Antibody Engineering, and they have the following tools for sequence analysis: Sequence storage and powerful querying, Configure workflows and automation by API and a Secure cloud platform.
Developing antibody therapeutics is challenging since they need to exhibit desirable qualities in two main domains: target (antigen) binding and developability.
In particular, developability is a wider term that is commonly used to refer to safety, manufacturability and storability of the antibody, and is measured by quantifying several physicochemical and biophysical parameters of the antibodies including solubility, stability, immunogenicity, structural dynamics and aggregation.
Accordingly, researchers have been developing high-throughput computational tools that can predict or calculate in silico the values of antibody developability parameters for large datasets of antibodies. These computational developability tools can be broadly divided into two main categories for:
Sequence-based developability prediction that requires the sole knowledge of the amino acid sequence of the antibody to predict developability parameters (DPs) values. The input for such tools is usually in the form of simple sequence strings or a FASTA file. Examples of these tools are BioPhi for immunogenicity prediction and sequence humanization and SoluProt for solubility estimation. And
Structure-based developability prediction that requires the knowledge of the antibody structure. This can be in the form of the experimentally defined (crystal) structure or in the form of the predicted model of the antibody molecule. In both cases, the input is usually in the form of a pdb file, but also the newer CIF format. Examples of these tools are FreeSASA for calculating solvent accessible surface areas and PROPKA for calculating the molecular electrochemical properties such as charge heterogeneity and 3D-based isoelectric point.
Some of these tools offer user-friendly interfaces via web servers such as the therapeutic antibody profiler (TAP) for the assessment of overall developability in relation to clinical stage antibodies, and CamSol for solubility prediction and sequence optimisation.
The overall complexity of this process encouraged the implementation of ML to develop models that can learn the underlying patterns of antibody sequences in order to predict the values of high-level DPs.
For example, netMHCIIpan is an ML model that uses artificial neural networks (ANNs) to predict the immunogenicity of proteins, including antibodies, starting from sequence input. This model was trained on experimental peptide-binding measurements to MHC II molecules which majorly reflects immunogenicity. Another example is SSH2.0 which uses a support vector machine-based (SVM) ensemble model trained on experimental data from 131 antibodies, to predict hydrophobic interaction risk of antibodies using sequence input.
However, as experimental developability data is scarce and the number of clinically approved antibodies is considered small (few hundreds), it is challenging to generalize these ML models on new antibody candidates.
📢 From PipeBio’s blog: Source
Founded in 2020 by Jannick Bendtsen, co-founder and CEO and Owen Bodley, PipeBio now has an ambitious team behind that comes from different backgrounds, ranging from wet-lab biologists to computer engineers, with both academic and industrial experience.
⚙️ Pepper Bio
Pepper Bio founded in 2018 in US 🍔 by Jon (Yu) Hu (CEO & co-Founder, MBA, Harvard Business School) and Samantha Dale Strasser (CSO & co-Founder, PhD, MIT) is a transomics company integrating phosphoproteomics, proteomics, transcriptomics and genomics, in a whole-stack solution 🕸️ for drug and biomarker discovery. By fully stacking and integrating these four layers, researchers have a complete, real-time map of what happens in cells before and during disease.
However, phosphoproteomics is very time-sensitive. So, not only is phosphoproteomic data typically hard to interpret but obtaining high-quality, meaningful data is tricky, and a lot of public data is unreliable since samples aren’t treated consistently. That’s why Pepper is building up the world’s largest transomic database of reliable phosphoproteomic data with corresponding proteomic, transcriptomic and genomic data, all from the same biological samples.
With their transomics technology, they have already demonstrated proof-of-concept across the three disease categories: neurodegenerative disease, oncology and inflammatory disease.
Pepper Bio has raised $6.5M in seed funding led by NFX (a venture firm exclusively focused on pre-seed & seed stage startups).
⚙️ Osmo
Osmo, founded in 2023 in Cambridge, Massachusetts 🇺🇸 as a spinout from Google Research by Alex Wiltschko, uses ML and has created a map of odor giving computers a sense of smell to improve the health and wellbeing of human life.
To digitize a sense, you must be able to
Read it and reading requires converting atoms into bits like what a camera does for light and a microphone does for sound,
Map it and mapping the sensory world involves understanding and organizing those bits like RGB for color and frequency for sound, and
Write it with distinct yet interdependent technologies and writing a sense means turning bits back into atoms that we can perceive like we do with printers and speakers.
They are now embarking on end-to-end reproduction of a captured scent and they call this an Osmograph, and believe that like a photo or a song an Osmograph has the power to activate treasured memories and evoke profound emotions.
More specifically, in order to find out if a computer could predict smells 👃 based on molecular structure they built a ML model using graph neural networks and trained the models using a dataset of 5,000 known compounds that were paired with the smell labels they evoked, such as "fruity," "floral," “cheesy,” or "minty." Then they froze the model in place and selected and tested it with 400 compounds. The idea was to see how the model predicted the scents of truly new molecules compared to 15 human panelists. Facing this maximally difficult challenge, their model made perceptual predictions that matched the empirical data 50% of the time, vs <20% of the time for the baseline model.
Osmo raised a total of $63.5M and investors include Bill & Melinda Gates Foundation, Lux Capital, Google Ventures, Google Research, Moore Strategic Ventures and 12 more.