How a Fortune pharma giant successfully used 4000 machine learning models and 120 AI workflows to accelerate drug discovery, streamline clinical data management, and improve productivity by 50% in the wake of the COVID-19 pandemic
It has been nearly 60 years since the publication of Claude Shannon’s landmark paper ‘A Mathematical theory of Communication’ that established the field of Information theory. Pharmaceutical companies, in particular, have benefited from being able to transfer information via digital networks, the proliferation of multiple external mnemonic devices that both collect and store information has given rise to a new problem: information overload. As human cognitive capacities are being stretched to the limit, a new method of information processing is required.

In response to this need, we worked with the client to develop the Neural Embeddings for Language Processing (NELP) platform. This was a homegrown variant of the standard NLP platform. Our team of experts built this platform that applies NLP, automation, and machine learning in pharmacovigilance. End result? Faster clinical trials, safer patient outcomes, improved employee productivity, and improved regulatory compliance.

The Challenge
High complexity of the case intake in pharmacovigilance process
When it comes to processes that are unique to the client, there are a series of challenges to solve for. For starters, automating the case intake process and the subsequent process of safety cases is all the more complicated in the face of the eudravigilance system that requires procedures to be nonserious and non-interventional. Other, more niggling issues surround the feasibility of safety signal processing, routine causality assessments, and the need for deriving publication-specific expression analytics.
Data mining to derive useful clinical trial information
Despite exposing the training models to GxP data, there is a slew of problems associated with mining and processing the relevant information. For instance, the feature of processing the investigational brochure to enable the first phase of in-human decisions poses distinct computational problems. Other pharmacovigilance concerns involve effective literature case processing, and automating the safety case triaging process, assistance with target liability assessment and the mining of relevant clinical trial information.
Delivering clinical summary during the prevention phase of pharmacovigilance exercise
In the prevention phase of any pharmacovigilance exercise, a robust NLP model ought to be able to provide mining, prediction, and report generation services. Specifically, the mining of associated patent information, molecular structure prediction, and the automation of statistical reports are expected to be dealt with thoroughly and effectively. Finally, all of these ought to be taken apart to put together a clinical summary and deliver an automated narrative.
The Solution
Build efficient systems using digital to enable drug discovery at scale
For our client, we began with the objective of putting together a scalable discovery engine by applying advanced computing techniques to existing data science practices. We were also required to connect therapeutic areas to all the strategic capabilities. This was then followed by unlocking development systems meant to boost the process of clinical trials while improving the automation and monitoring capabilities of regulatory/safety systems, while evolving a robust data fabric. Finally, our system engineering efforts culminated in defining the first platform with real-world API collaboration and optimized scientific computing.
Address the operational and people-related concerns
While the NELP did focus on solving the computational problems that surfaced in the attempt to expedite information processing, we were careful not to neglect the impact it could have on the operational aspects of the organization. Early on, we developed a community that would be proficient in IP practices pertaining to data, open-source, and patent protection. We also made sure to impart design skills throughout the organization to align diverse talent with the objectives at hand.
Stay Ahead
Visit our Pharmacovigilance page
The Impact
Smarter pharmacovigilance operations to quickly navigate through the pandemic
The NELP platform was developed at an opportune time to help our client with their pharmacovigilance practices. As the COVID-19 pandemic continues to evolve with increasing speed, thereby prompting a rapid response from pharmaceutical organizations, it was absolutely essential to leverage every available automation technology to accelerate the pace of breakthroughs and optimize operations within the medical research community. The only real way to do that is by employing intelligent systems that can offload some of the work from human researchers, thereby enabling the latter to focus on synthesizing solutions.
Surprisingly, the NELP platform’s capabilities granted it a degree of versatility that we hadn’t initially expected. As the product combines analytics, machine learning, and natural language processing along with the ability to deliver at scale, it found use cases that transcended all aspects of pharmacovigilance. It has helped with Eudravigilance, legal cases, discovery sciences, target liability assessments, consumer call centers, and a slew of other areas.
Liked this transformation story?
Let's build yours now.
These wide-ranging applications means that the NELP is more than just a tool, but a fully integrated enabler with transformative effects that have enhanced the organization in nearly all aspects. Such an achievement was sorely needed in light of the current state of clinical teams that work round the clock to contain the COVID-19 pandemic.
Effective lead discovery for high-value biological insights
The success of any given trial begins with a trouble-free detection exercise. Identifying the kind of leads that would necessitate work later on means that teams need the potential to extract high-value biological insights. This needs to be done at scale during emergency situations such as a pandemic. Our NELP platform enables this via high content imaging and transcriptomics technologies, thereby helping teams succeed in effective lead discovery.
Efficient data mining of medical literature for effective drug and clinical insights
Once meaningful leads are generated (and this number can run in the thousands, if not millions), teams are faced with yet another dilemma that has historically contributed to the slow pace of drug development: target selection. The only constraint here is processing abilities, and this can be remedied by the use of our NELP platform. It facilitates the mining of very large data sets of medical literature, enabling the discovery of the right target, molecule, and clinical plan.
Accelerated drug discovery and productivity improvements of 50%
The NELP platform’s integrated machine learning operations solution succeeded in reducing the cycle time while concomitantly improving the NME quality across all therapeutic areas. This translated to the enablement of 3.4 billion compound predictions that were a result of over 1000 machine learning models and over a 120 high-scale AI workflows. These combined results also contributed to a whopping 50-60% increase in the productivity of medicinal chemists.
Touchless decision making
Going beyond the detection and selection face, we also helped our clients in the area of expediting their first-in-human decisions through our touchless data integration solution. This support was also extended to advanced tox analytics for therapeutic areas, non-clinical studies as well as safety scientists. NELP was also leveraged for COVID-19 analytics on global published literature and sparked a vaccine digitization and analytics transformation program.
Robust risk management of clinical trials
For purely clinical solutions, our NELP platform was able to leverage the COVID-19 clinical ops risk predictor, and along with SIROP, our AI solutions enabled the reinstating of over 500 global clinical trials. This was made possible due to the generation of over 4000 machine learning models in the course bringing tangible business impact.
Improved compliance in pharmacovigilance and regulatory affairs
As clinical trials continue to be heavily subject to eudravigilance norms, our NELP platform eased our clients’ regulatory burdens by successfully executing automated HA intakes, IDMP extractions. This level of automation was also extended to country-level submissions by leveraging the platform’s MLOPS and NLP capabilities. Through these capabilities alone, we were confident in establishing the realization of a $2.5MM roadmap for 2021.
High-volume of accurate literature case processing in pharmacovigilance
Our NELP platform’s literature automation tool has processed more than 250,000 different pieces of medical literature to predict a grand total of 13,247 cases with a staggering 97% accuracy. Doing this for our client unlocked north of $2MM value via cognitive automation on literature, Eudravigilance, NSNI cases, consumers, Bolus, and business partner cases. The present level of automation alone touches over 100,000 - 150,000 cases.
AI, machine learning, automation, and NLP for pharmacovigilance transformation
AI, machine learning, automation, and NLP for pharmacovigilance transformation