June 17, 2019 –
Several advances in A.I. are making a transformative impact on drug R&D—changing the landscape and making it possible to finally uncover the molecular underpinnings and complex pathways of diseases that have long remained mysteries.
As covered in Deep Knowledge Analytics’ recent report:
“Discovering new drugs using A.I. is one of the most challenging areas in biological sciences. Top tier A.I. for Drug Discovery companies have distinguishing characteristics that include high levels of expertise in biopharmaceutical science, advanced proficiency in A.I., very specialized teams, and constantly evolving internal knowledge.
Although there are about 150 A.I. companies in the Drug Discovery space, very few of them are capable of building end-to-end solutions.”
Deep Knowledge Analytics went on to name WuXi NextCODE as one of the few leaders able to provide end-to-end A.I. in drug R&D, echoing FastCo’s own Top 10 list.
The chief of WuXi NextCODE’s A.I. program is Tom Chittenden, PhD, DPhil, PStat, who is a Forbes top 100 “pioneer” of A.I. in drug development.
Dr. Chittenden, who earned PhDs in both Molecular Cell Biology and Computational Statistics, and holds appointments at Boston Children’s Hospital and Harvard Medical School, has a long list of works in Nature, Science, and other publications detailing A.I. approaches to uncover disease pathways and drivers.
One such example is the recent publication Endothelial ERK1/2 signaling maintains integrity of the quiescent endothelium that provides experimental validation for the novel WuXi NextCODE A.I. strategy for the life sciences.
WuXi NextCODE’s ensemble A.I. pipeline is unique in its ability to integrate multi-omic data sets to understand underlying biological context. Traditional A.I. approaches search for correlations between data points—such as between a gene variant and a disease phenotype or between gene regulation and disease severity—looking for areas of overlap to infer a potential disease pathway. But these approaches can provide black box predictions that fail to provide causal insight into a set of biomarkers that have emerged at the top of the list. The WuXi NextCODE A.I. approach, coined phenotype projection, allows for predicting complex phenotypes by casting light on the causal molecular underpinnings of disease. This method combines the use of ensemble deep learning with probabilistic programming that applies Bayesian models to test the causal dependencies of millions of possible interactions between proteins to elucidate the core biological network connecting the observed phenotype with the mechanism of disease. This has been validated most recently in a publication in JEM where phenotype projection was applied to map out the complex communication between ERK1/2 and TGFb pathways and resultant disease states of renal failure, hypertension and sudden death from myocardial fibrosis. The causal gene network accurately predicted the observed phenotypes, including a specific cause of the observed systemic hypertension phenotype and renal dysfunction. WuXi NextCODE applies a proprietary statistical machine learning approach to identify causal genes and molecular pathways, representing putative drivers of disease.
One of the advantages of WuXi NextCODE’s offering is our access to large, disease-specific, well-curated data that includes multi-omics from tissues, whole genome sequencing, and a very deep phenotype. Through its relationships with international healthcare providers, WuXi NextCODE is able to assemble data on thousands of patients and controls for over 60 different diseases for analysis. The largest of those partners, Genomics Medicine Ireland (GMI), is sequencing (WGS) 400,000 patients and controls for 60 diseases in the homogeneous population of Ireland.
All this data is curated within WuXi NextCODE’s global CLIA/CAP labs and stored in a proprietary and secure genomics platform allowing partners to easily apply A.I. and statistical analysis algorithms to the data, testing unlimited hypotheses through stratification of the data based on phenotype and genomics—with a single analysis taking only minutes to hours to conduct and report on.
In addition to the probabilistic programming to discern causal relationships, WuXi NextCODE’s ensemble A.I. approach is transformative for a number of other reasons.
Feature Learning to Prevent Information Loss
Firstly, Dr. Chittenden’s team has developed novel feature learning strategies to integrate and reduce complexity of high-dimensional multi-omic and phenotypic data without the loss of information content. This is especially important when deriving informative, reproducible disease-specific information from high-throughput sequencing technologies, such as whole genome DNA-seq data, as these data associate with high degrees of feature dependencies and correlation bias.
Retrosynthetic Reaction Prediction
The WuXi NextCODE A.I. applications also extend to chemical synthesis planning for drug discovery. This data-driven approach for retrosynthetic analysis applies deep highway networks in two steps: (1) first predict a group of reactions to produce a molecule and (2) subsequently predict the transformation rule that produces a molecule within the reaction group. This work is highlighted in the recent publication, Enhancing retrosynthetic reaction prediction with deep learning using multiscale reaction classification (JCIM) that showed a novel multiscale, deep highway network approach to successfully generate valid reactants from retrosynthetic reaction predictions.
Quantum computing promises to significantly advance many subfields of computational science, including statistical computing and computational biology. Anticipating its utility in the biomedical sciences, we implemented several quantum statistical computing strategies on a simulated universal quantum computer and a physical quantum annealer. In Dr. Chittenden’s recent keynote at the 2019 Qubits Europe conference in Milan Italy, he presented a novel quantum machine learning strategy that outperformed several classical statistical computing methods on small patient datasets—a feat that is anticipated to revolutionize the biomedical sciences. Furthermore, Dr. Chittenden’s team demonstrated that similar performance can be attained with simulated quantum machine learning, based on a statistical physics method without the use of quantum computers. This means drug developers can create big value with small patient datasets to identify causal biomarkers of disease, repurpose or reposition drugs, and discern molecular signatures of drug response in patients.
Interested in learning how this might benefit your own analysis objectives? Contact WuXi NextCODE at firstname.lastname@example.org.