Neural Information Processing Systems Conference (NeurIPS 2018)
BenevolentAI’s team will be attending and speaking at NeurIPS 2018. We look forward to meeting you there! We will be at the exhibitor area on booth number 615. If you’d like to get in touch before or during NIPS contact us!
Sunday 2 December at 16h30, the BAI team will be giving a 60 mins talk about our AI platform and discuss the great opportunity it represents for pharmaceuticals and drug development.
Artificial Intelligence for Target identification in Drug Discovery with Amir Saffari, VP AI research at BenevolentAI.
A key challenge in the drug discovery process is the identification of potential therapeutic targets for a given disease. This process requires selecting both a target and an entity, often a compound, for modulating the target’s activity to validate its association with a disease. An additional complication is that heretofore unknown targets may be more challenging to identify but offer increased opportunities for the development of novel drugs. At BenevolentAI, we use machine learning throughout the entire process. Relation extraction models drive our pipeline. In conjunction with structured data and our own experimental results, this processing pipeline ingests information from scientific publications, abstracts, patents. Next, this biological knowledge graph is leveraged to form predictions using relational inference algorithms, including matrix factorisation and graph convolutional models. These predictions are aggregated together with machine learning models built on genomics data and information mined from text. Druggability, tissue specificity, and other metadata are then added to surface the most promising targets to test in the lab. This talk will describe our process and highlight some of our early successes .
ML4H: Machine Learning for Health Workshop
Our team will present posters on Saturday 8 December at the Machine Learning for Health workshop.
This workshop will bring together machine learning researchers, clinicians, and healthcare data experts. The program consists of invited talks, contributed posters and panel discussions.
> Graph Convolutional Networks for Inference on Noisy Knowledge Graphs (Daniel Neil, Aaron Sim, Joss Briody, Alix Lacoste, Páidí Creed, Amir Saffari)
We provide a new formulation for Graph Convolutional Neural Networks (GCNNs) for link prediction on graph data that addresses common challenges for biomedical knowledge graphs (KGs). We introduce a regularized attention mechanism to GCNNs that not only improves performance on clean datasets, but also favorably accommodates noise in KGs, a pervasive issue in real-world applications. Further, we explore new visualization methods for interpretable modelling and to illustrate how the learned representation can be exploited to automate dataset denoising. The results are demonstrated on a synthetic dataset, the common benchmark dataset FB15k-237, and a large biomedical knowledge graph derived from a combination of noisy and clean data sources. Using these improvements, we visualize a learned model’s representation of the disease cystic fibrosis and demonstrate how to interrogate a neural network to show the potential of PPARG as a candidate therapeutic target for rheumatoid arthritis.
> A Comparison of Methods for Progression Endotype Detection in Amyotrophic Lateral Sclerosis, Hamish Tomlinson, Romain Studer, Poojitha Ojamies, Joanna Holbrook, Paidi Creed
The discovery of new treatments for amyotrophic lateral sclerosis (ALS) is challenging due to the heterogeneity of disease progression in patients. Characterising distinct patient endotypes may improve patient stratification, thus enabling more effective clinical trials. We present ALS endotype discovery as a time-series clustering problem and compare progression endotypes found using a probabilistic subtyping model (PSM), a K-means based model and a dynamic time warping (DTW) model. We used the elbow method to find the number of endotypes to be 5 for all models. Pairwise log-rank tests showed that most of the discovered endotypes differed significantly with respect to survival. Adjusted rand index analysis revealed that the PSM and K-means models found similar clusters. Overall, we discovered more distinct ALS endotypes than widely believed to be present and demonstrated differences and similarities between time-series clustering approaches
We are inviting you to our booth where we’ll show you demos of some of our AI and machine learning tools that help our scientists to accelerate our journey from data to medicines.
Our approach is to integrate technology across the entire medicinal R&D process (rather than fragments of it) and by doing so solve fundamental innovation roadblocks such as the ability to ingest large data sets, meaningful reasoning on those data sets, validation of ideas from those data sets and the rapid experimentation of those ideas.
Come and say Hi to us on booth 615, chat with our team about job opportunities at BAI, or discuss possible collaborations and partnerships.
Join the team
If you asked our team why they have joined BAI, most of them will respond: “What we do truly matters”. Our collaborative approach with our in-house drug discovery and development scientists bring our technology at the closest of our scientists needs. Our aim? Bring better treatments for every patients.
Paidi, one of our senior ML researcher says:
“At Benevolent I can work on interesting and challenging problems, and see my colleagues apply the solutions I develop in real drug discovery programmes as soon as they are available.”
If you aspire to truly make a difference, it is probably the time to cease the opportunity and find out more about how you can contribute. Come and meet us!
In the meantime, here are a few open positions in London and New York that you may be interested in:
If you are not finding any jobs suitable for you, please reach out anyway. We are looking for brilliant minds, always!