Accelerating Biomedical Innovation with NLP and Knowledge Graphs at Roche Information Solutions

In the ongoing pursuit of advancing healthcare, the relentless drive to innovate with cutting-edge technologies persistently transforms the expansive domain of biomedical research and development. At Roche Information Solutions, this pursuit takes a transformative turn through the fusion of advanced Natural Language Processing (NLP) and the structured intelligence of Knowledge Graphs.

This case study delves into how Roche harnesses the power of these sophisticated tools to accelerate biomedical innovation, streamline processes, and redefine the boundaries of scientific discovery in healthcare.

About Roche Information Solutions

Roche Information Solutions, part of the Roche Diagnostics family, stands as a driving force in pushing the boundaries of medical progress and revolutionizing diagnostics and treatments for patients globally. As a John Snow Labs customer since August 2018, Roche Information Solutions embodies the philosophy of “Doing now what patients need next.” This commitment is ingrained in its two divisions, Diagnostic and Pharmaceutical, both dedicated to delivering meticulously tailored and personalized care to patients worldwide.

The journey of Roche Information Solutions began as an initial Proof of Concept (POC) that sought to validate the knowledge graph concept using solely public data sources. Evolving progressively, it transitioned from information retrieval-based use cases to a pivotal focus on Biomarker Discovery. Roche is currently in the process of publishing a peer-reviewed journal article that showcases the transformative potential of Link Prediction for Biomarker Discovery, signifying a remarkable milestone in the company’s pursuit of healthcare advancement.

With a history spanning over 125 years, Roche is committed to gaining insights that improve patient care and enhance our understanding of diseases. Teaming up with Wisecube, Roche Information Solutions is pioneering the utilization of NLP and knowledge graphs. This collaboration is changing the game in biomedical research, making big strides in patient care—a major step forward in healthcare evolution.

The Challenge: Navigating the Biomedical Data Explosion

Biomedical data management and exploration presents several critical challenges posing significant obstacles to leveraging valuable knowledge from vast volumes of unstructured textual data:

  • Data Explosion: The exponential growth of biomedical literature, doubling every three years, presents an insurmountable obstacle for individuals or resources to keep pace with.  This surge in information, inclusive of publications, clinical trials, and research papers, presents a monumental challenge in efficiently managing and extracting meaningful insights.
  • Data Fragmentation: The landscape is riddled with fragmented, small-scale datasets, hindering the synthesis of comprehensive insights. This fragmentation obscures potentially groundbreaking discoveries and innovation within the field. There’s a pressing need to interconnect essential pieces of biomedical information, facilitating easy access and interrogation with specific queries. This demand for linking disparate yet crucial data sets looms large in an environment inundated with unstructured text.
  • Data Access Barrier: An astounding 85% of biomedical research findings are inaccessible for replication due to limited data availability. This lack of access hampers the validation and advancement of critical discoveries.
  • Data Bias Amplification: Siloed biomedical data exacerbates the prevalent bias issue in healthcare research. Isolated datasets contribute to perpetuating biases, impeding impartial and comprehensive insights.

Recognizing these challenges, Roche sought to tackle the formidable task of unlocking and distilling pertinent information from this deluge of biomedical data. The goal was to extract, organize, and synthesize key insights, enabling informed decision-making within the realm of healthcare.

Wisecube’s Solution: Leveraging NLP & Knowledge Graphs For Contextualizing Biomedical Insights

Wisecube came through with just the right tools for Roche Information Solutions to revolutionize biomedical data management. It offered a robust platform for Roche to leverage the integrated powers of NLP and knowledge graphs to overcome the challenges of contextualizing biomedical information.

The emphasis on contextual AI fosters a human-centric approach to AI, ensuring that the derived biomedical insights are not only comprehensible to AI systems but also easily understandable and interpretable by human users. The aim of this approach is to facilitate more effective and collaborative decision-making processes in healthcare.

Architecture of Wisecube’s NLP-powered Knowledge Graph Solution (Source: Wisecube)

The partnership project between Wisecube and Roche progressed through several vital stages to implement the contextual AI approach:

Data Collection and Normalization 

The project initiation involved gathering a myriad of structured and unstructured public data sources, including Wikidata, PubMed, Clinical Trials databases, and ChemBL. This diverse array of datasets underwent meticulous normalization into a unified model, facilitating seamless integration and compatibility. The normalization process not only ensured harmonization but also rendered the data searchable and easily accessible for efficient navigation and information retrieval.

A simplified schema of the Biomedical Knowledge Graph

Information Inference Using NLP

The next step involved focusing on uncovering the inherent meaning embedded within unstructured datasets. Leveraging John Snow Labs’ healthcare and biomedical NLP models, the system excelled in extracting over 30+ named entities and relationships from the text. The inference process entailed identifying and deciphering crucial relationships among entities such as diseases, genes, and drugs. The inferred structural patterns from these entities and their relationships enabled the extraction of nuanced information vital for comprehensive biomedical understanding.

Construction of a Knowledge Graph

Drawing upon graph thinking principles, Wisecube constructed a robust knowledge graph to conceptualize the healthcare domain within a connected graph structure. This step involved unifying and interlinking disparate biomedical facts within a comprehensive graph, using existing ontologies to enrich its context and relevance. The resulting graph framework encapsulated over 5 billion semantic facts, enriched with additional data to bridge gaps within existing datasets. Furthermore, the knowledge graph utilized advanced graph algorithms like link prediction and node classification to enhance its predictive and analytical capabilities.

Graph Validation and Enhancement

The next step encapsulated validating the knowledge graph to evaluate its effectiveness as a reliable biomedical knowledge framework. The graph was evaluated against over a thousand biomedical queries. This also involved continuously improving the graph’s accuracy and relevance by incorporating additional datasets and refining the underlying model.

Knowledge Discovery Through Intuitive UI

The next stage was aimed at democratizing access to the knowledge graph and facilitating information discovery. An intuitive interface was crafted to enable users across varied expertise levels to seamlessly search, pose inquiries, and navigate through the graph without necessitating technical proficiency in querying languages. To enhance biomedical context discovery, the graph can be integrated with personalized analytics and tools to enable insights tailored to specific organizational needs.

Outcomes: Wisecube’s Solution Achieved 82% Recall Rate in Biomedical Inquiries

Wisecube’s knowledge graph solution demonstrated an impressive 82% recall rate when subjected to over 3240 natural language biomedical queries sourced from the BioAsq training dataset. This high recall percentage underscores the solution’s efficacy in efficiently and comprehensively addressing a wide spectrum of intricate biomedical inquiries.

The strategic implementation of Wisecube’s solution framework empowered Roche to cultivate a robust biomedical knowledge ecosystem, opening doors for a diverse spectrum of users. This partnership not only enabled Roche to promote informed decision-making but also propelled innovation within the healthcare domain, aligning with its overarching goals.

The integration of NLP and knowledge graphs at Roche yielded several impactful outcomes:

  • Enhanced Data Accessibility: Researchers and healthcare professionals are now able to efficiently comprehend and leverage healthcare insights using the enhanced accessibility to vast biomedical datasets.
  • Improved Decision Making: Integration of disparate data sources into a cohesive knowledge graph facilitated more informed decisions in the development of medications and diagnostics, enhancing precision and efficacy.
  • Time Efficiency: Significant reduction in the time required to stay updated with the latest research allowed healthcare professionals to allocate more focus and resources towards patient care.
  • Scalability: The project showcased the scalability prowess of combining NLP and knowledge graphs to manage and derive meaningful insights from extensive biomedical data, demonstrating its adaptability to large-scale applications.

Pioneering the Future: NLP and Knowledge Graphs in Shaping Healthcare

Roche Information Solutions, working hand-in-hand with Wisecube, highlights the incredible potential of NLP and knowledge graph technologies within the biomedical ecosystem. This dynamic duo brings unmatched abilities: NLP’s knack for deciphering complex biomedical text and KG’s seamless weaving of scattered medical info, all while embracing the principles of contextual AI. The synergistic effects of these technologies are not only limited to streamlining healthcare research, they are also setting the stage for revolutionary advancements in personalized patient care. 

In addition to being a success story, this case study serves as a benchmark for similar ventures in healthcare, encouraging the incorporation of AI to boost health outcomes and drive ongoing improvements. To delve deeper into Wisecube’s transformative collaboration with Roche, tune into our enlightening NLP Summit webinar featuring Dr. Antoeneta.

Table of Contents

Scroll to Top