AI/ML tools, startups and investors for preclinical drug discovery (2nd part)
An overview of AI/ML startups 🚀 and investors transforming preclinical drug discovery
Pharmacogenomics, startups and investors for preclinical drug discovery
For the first part of this newsletter: 💠 AI/ML tools and startups for preclinical drug discovery.
➡️ Artificial Intelligence And Preclinical Studies
Once a lead compound has been identified and validated during drug discovery, the preclinical phase is initiated for testing drug efficacy and safety in various disease models. As a result of the accumulation of all the different preclinical disease models—such as patient-derived cell lines, organoids, xenografts, rodent models, zebrafish models, organs on a chip and cell cultures—together with the advances in computational approaches for drug response prediction, pharmacogenomics has emerged investigating the role of genetic events in drug responses.
To make a long story short, as a consequence of all the large-scale pharmacogenomic screening experiments of the preclinical disease models assisted by the omics technologies—i.e. genomics, transcriptomics, epigenomics and proteomics—applying in parallel computational methods—Network-based prediction models, ML models and DL models—a number of FDA-approved pharmacogenomic biomarkers have emerged, namely a measurable DNA and/or RNA characteristic that is an indicator of normal biologic or pathogenic processes, and/or response to a therapeutic. The list of all the FDA-approved pharmacogenomic biomarkers (up to 2020) in the labeling of anti-cancer drugs can be found here: Large-scale pharmacogenomic studies and drug response prediction for personalized cancer medicine. While some representative DL models used during pharmacogenomic screening for drug response prediction are the following:
DNN model (A Deep neural networks trained on database of 1,001 cancer cell lines),
CDRscan (A dual convergence architecture DL model, which predicts somatic mutation profile-based drug responsiveness based on a large-scale drug screening assay data encompassing genomic profiles of 787 human cancer cell lines and structural profiles of 244 drugs),
MOLI (A multi-omics late integration method based on DL networks),
tCNNS (A twin Convolutional Neural Network for drugs in SMILES format),
PaccMann web server (A multi-modal attention-based neural networks trained on public transcriptomic cell line profiles, compound structure information and drug sensitivity screenings) and
DrugCell.
In particular, the DrugCell is a “visible” neural network that predicts anti-cancer drug responses by modeling the hierarchical organization of a human cancer cell. DrugCell uses a modular neural network design that combines conventional artificial neural networks (ANN) with a visible neural network (VNN) to make drug response predictions. Binary encodings of individual genotypes are processed through a VNN with architecture guided by a hierarchy of cell subsystems, with multiple neurons assigned per subsystem. And compound chemical structures are processed through an ANN using the Morgan fingerprint as input features. It was trained on 1,235 tumor cell lines and their responses to 684 FDA-approved and experimental therapeutic drugs, and some of DrugCell's conclusions were validated in laboratory experiments.
➡️ Startups Using AI Models During Preclinical Studies
✅ A UK and Australia-based startup, Concr, allows cross-talk between biopharma, diagnostics companies and healthcare providers by unifying data on a single platform to address meaningful questions in precision oncology. Its platform FarrSight® integrates diverse data to generate actionable insights, ranging from biomarker discovery during preclinical development to patient outcome prediction in clinical trials. Concr founded in 2018, is utilizing a ML algorithm to improve prognosis of patients with cancer of unknown primary, by aiding diagnosis of tissue of origin, while a DL model was developed to reliably identify 3 cell types relevant for treatment of triple negative breast cancers using H&E-stained slides, with performance validated by an expert pathologist. On August 16, 2023, Concr welcomed Oncology Ventures as investors in the second tranche of their oversubscribed seed round (a total of. £2.9M in funding over 3 rounds).
✅ Vivodyne (US, 2020) enables human testing before clinical trials with AI‑scale testing on lab-grown human organs. With over 20 human organ models of health and disease that span the major organ systems, their models mimic native human functions & phenotypes to accurately capture the effects of new therapeutics and predict clinical outcomes. In particular, each of their proprietary Data Engine robotic machines tests on 10,000+ independent human tissues at a time, yielding vivarium-scale output in the footprint of a lab bench. Then their AI platform is powered by an active learning system that can identify and produce the most useful human training data on-demand to refine its accuracy. By integrating data from 3D phenomics, single-cell sequencing and proteomics Vivodyne deeply understands and maps human complexity. Back in November 2023, Vivodyne raised $38M in total seed financing, led by Khosla Ventures with participation from CS Ventures, Bison Ventures, MBX Capital and Kairos Ventures.
✅ On October, 11, 2023, Antiverse and GlobalBio have extended their collaboration to advance antibody cancer therapeutics. In particular,AntiverseHQ(UK, 2017), a biotechnology company utilizing ML to design antibodies against difficult targets such as GPCRs and ion channels, and GlobalBio, an antibody engineering company developing methods to engineer improved and more developable therapeutic antibodies, announced that they will be extending their collaboration to advance immune checkpoint inhibitors in cancer therapy. The initial collaboration successfully resulted in the generation of a panel of anti-PD-1 antibodies (programmed death-ligand 1 or PD-L1 also known as cluster of differentiation 274, CD274, has been speculated to play a major role in suppressing the adaptive arm of immune systems) with diverse binding and functional profiles, with two candidates from this panel now entering preclinical development.
Antiverse’s proprietary AI-driven Antibody Discovery platform, uses state-of-the-art ML techniques and advanced cell line engineering to develop antibodies for challenging drug targets, alongside GlobalBio’s ALTHEA semisynthetic libraries for the discovery and optimisation of antibody-based therapeutics. Since its is difficult to discover antibodies against GPCRs receptors because of cells with low receptor count and because of antibody libraries with poor specificity, at Antiverse they have built a discovery platform that develops cells with high receptor counts (1M per cell) and antibody libraries (de-novo target-specific libraries focused on functional epitopes) with high specificity against each target. Subsequently, in their lab they screen the target-specific libraries against cells with high receptor counts to reach a high number of binders and the outputs are then sent for deep sequencing and AI-based clustering. Antiverse has raised a total of £4.3M. in funding over 8 rounds.
✅ Neuron23 (US, 2018) is combining cutting-edge data science and human genetics, to create precision treatments to fight debilitating diseases.
Back in 2022, Qiagen and Neuron23 entered an agreement for the development of next-generation sequencing (NGS) companion diagnostic for Neuron23’s brain penetrant leucine-rich repeat kinase (LRRK2) inhibitor for treating Parkinson’s disease. The collaboration between the companies will also support the clinical development of Neuron23’s drug candidate, which is now in the late phases of preclinical development. Neuron23 has raised $213.61M to date.