AbbVie, Discngine Collaborate On Structure-Based Drug Discovery Explorer

June 26, 2019

By Allison Proffitt

June 26, 2019 | To be honest, they weren't sure they could do it as it was a massive undertaking. And in a "fail early fail safe" world, Rishi Gupta, Principal Research Scientist, AbbVie, had a few months to show that this would work and proceed to build it up.

The Structure Biology group at AbbVie had been collecting protein structures for years, Gupta explains. About four years ago, it became clear that AbbVie had all these structures, "but there was no consistent way for us to access those and provide the value out of that to our scientists within the medicinal chemistry organization," he says. AbbVie scientists couldn't exploit the structural knowledge to its full potential and the decision-making process in structure-enabled drug discovery projects was slow and cumbersome.

Meanwhile, Gupta was already working with Discngine on a cheminformatics platform. "Peter Schmidtke [product manager for Discngine] and I were talking about all of these cool concepts and we discussed that we should probably work on an application where we could expose all of these protein structures, visualize them, and then go from there," Gupta remembers. "I knew they had the capability to help us; it was a match made in heaven due to prior success."

The main objectives of the project were to centralize public and in-house structures in a unique, searchable database; to develop a web-based user interface that allowed non-experts to perform advanced analyses of the structural data; and to allow all scientists in structure-based drug discovery at AbbVie to create, capture, share, and apply new structural knowledge to their current and future projects.

Gupta says when he first presented the goals and a few rough drafts of the project, some on his team had doubts that anyone could take all of this information, pre-analyze and visualize it in a manner that could be consumed by the medicinal chemistry organization in a productive manner to extract all the knowledge. Gupta worked closely with the medicinal chemistry user group to set expectations and aimed for a proof-of-concept in a very short period of time. The development team then proceeded to work in an agile mode to continuously develop and improve the application.

The successful collaboration between Abbvie and Discngine—3decision—won a 2019 Bio-IT World Innovative Practices Award in April at the Bio-IT World Conference & Expo.

A Trusted Partner

Discngine, a Paris-based company, started about 15 years ago, but shifted to new generation software in about 2013, explains Eric Le Roux, CEO of Discngine.

For this project, Le Roux says, the Discngine team started "almost from scratch". Discngine already had some open source work created by some of our members here on druggable pockets. "For the rest of it, we made a proposal to Rishi and AbbVie to create from scratch an entirely new application."

The resulting product is 3decision—a decision-making tool for 3D structures. The team conceived a dedicated data model in order to properly organize and store structural data and metadata. This model was used to set up a fully annotated and searchable Oracle database.

The 3decision database brings together all of AbbVie's data sources along with public data:

  • In-house protein structures
  • Public protein structures
  • Protein and antibody sequences
  • Small molecule structures, fragments and ligand descriptors
  • The entire structurally resolved pocketome, pocket descriptors
  • Protein-ligand interactions
  • Statistics on protein-ligand contacts
  • Sequence and structure annotations
  • Structure-structure relationships
  • ChEMBL assay data
  • Project data
  • And more

Oracle is the relational database engine, and BIOVIA Direct serves as molecular cartridge. The rest of the backend is containerized and built up of BIOVIA Pipeline Pilot as processing backend and open source tools (e.g. fpocket, Rdock, rdkit). The web-based front-end was developed based on AngularJS and several JavaScript libraries. Several novel visualizations have been integrated using d3.js. Communication between front and backend is achieved through a REST API.

The tool is hosted on AbbVie's AWS EC2 (virtual private cloud instance), and the user interface gives AbbVie researchers direct access to advanced structural analytical tools and collaborative features. Both first-time users and the more experienced can perform large-scale structural analytics on the entire structural knowledge base. 3decision analysis and registration pipelines are available directly within AbbVie's internal LIMS, compound ideation platform etc.

"We were sitting on these almost 9,000 structures, and except for 2-3 people, nobody else knew where they were," Gupta says.

Having AbbVie's internal data explorable along with external data has improved the quality and speed of decision making in structure-based drug design. Scientists spend less time on tedious tasks like searching, gathering, and preparing structural data. Instead, they can use their time to create real value by generating new ideas that bring the project forward. And having historic data available from past efforts lets scientists extract and apply knowledge that has been generated over time by different project teams and research programs.

The collaborative features allow for simplified communication and project management.

Commercial Offering

The success of 3decision comes from the collaborative development approach, Le Roux emphasizes. "Instead of working in our respective garage and trying to reinvent the wheel, we had software developers working with scientists to create an innovative new product," he says. "We believe this is a way to do great R&D."

3decision is a commercially-available product now available from Discngine. Although not populated with AbbVie's in-house data, of course, the solution is otherwise identical, Le Roux explains.