The Bio-IT World Conference & Expo, taking place in Boston from April 21-23, is an annual event that brings together thousands of great minds to highlight technologies and analytic approaches to solving problems, accelerating science and driving the future of precision medicine.
Elsevier is proud to be on the ground there to showcase our collaborative projects, our high-quality Life Science Solutions and our expertise in data sciences. We’re happy that our own data scientist Finlay MacLean will be offering a presentation as a part of Track 11 (Bioinformatics) at the event. We hope you will attend! (Also be sure to swing by and visit us at Booth 348, as well!)
April 22 at 1:10pm
11 / Bioinformatics / Precision Cancer Medicine Methods
LUNCHEON PRESENTATION II: A Network Polypharmacology Approach to Diffuse Intrinsic Pontine Glioma
MacLean, MSc, Data Scientist, Elsevier
medicine promises to be a potential linchpin in oncological drug repurposing.
We developed a multi-scale heterogeneous knowledge graph spanning genomics,
epigenetics, transcriptomics and proteonomics. We implemented a random walk and
generated dense vector representations of the neighbourhoods (or interactomes)
of key nodes and used these in downstream supervised machine learning tasks.
Leveraging Entellect we plan to use the models in our collaboration with the
University of Zurich, to suggest potential DIPG drug repurposing candidates.
Here is a more detailed preview of his presentation:
network polypharmacology approach to Diffuse Intrinsic Pontine Glioma graph-based
machine learning on heterogeneous biological knowledge graphs for drug
indication and disease gene association prediction.
Diffuse Intrinsic Pontine Glioma (DIPG) is one
of the most aggressive malignant pediatric brain tumors. Chemotherapy has
proven to be ineffective, and surgical removal isn’t an option. Thus, hope only
lies in the development of effective therapeutics.
Whilst over 250 clinical trials have been
undertaken for DIPG, none have seen particular success. This is thought to be
in part caused by, i) the activity of efflux transporters in blood-brain
interfaces removing drugs before they can take effect, ii) the tumor
heterogeneity, and iii) the sparsity of regularly harvested tissue samples.
Network medicine (NM) promises to be a
potential linchpin in drug repurposing for cancer. NM aims to approximate the
complex intracellular and intercellular network and functional interactions,
exploiting the organising principles that govern such cellular networks to gain
greater understanding of disease. NM can be complemented with graph-based
machine learning methods, in which either i) the inherent structure of such
networks can be used directly to influence the network architecture of the
neural network, or ii) biological entities can be viewed as structural
elements, and the network topology can be learnt to determine correlations in
such structures. There is a strong synergy between polypharmacological (PP)
approaches to drug discovery and network medicine. PP strives to develop
therapeutics in which the totality of perturbations to the mechanistic pathways
of a disease restores the system to a healthy state. The pursuit of promiscuous
drugs; developed to target multiple pathways of a disease, is easily
facilitated with network medicine.
To implement a network polypharmacology
approach to DIPG, we developed a multi-scale heterogeneous knowledge graph of
proprietary and open-source databases such as ResNet (Pathway Studio), Reaxys
Medicinal Chemistry, OpenTargets and multiple drug repurposing and epigenetic
databases. In total, the graph comprised of 10 million edges (weighted by
occurrences in literature), and 1.3 million nodes. Using the knowledge graph,
we implemented a random walk (weighted by reference count) and generated dense
and accurate vector representations of the neighbourhoods (or interactomes) of
each disease, small molecule and protein node. An ensemble of supervised models
was used in conjunction with a confident learning regime to characterise
missing or invalid links within the network, notably novel disease gene
association and drug indication. Whilst the models were predominantly used for
detection of novel targets and small-molecule regulators, the model is also
capable of highlighting over-reported interactions and thus the potential
research bias inherent in literature-based knowledge graphs. Leveraging
Entellect we plan to utilise these models in our continued collaboration with
the University of Zurich, to suggest potential drug repurposing candidates for
this currently ‘untreatable’ disease.