Cancer is detected too late, but the innovation around multi-cancer early detection tests brings hope that we can change the detection paradigm and the outlook of a cancer diagnosis. This presentation will dive into the development of tools to catch cancer early and change how cancer is treated, and the importance of providing life changing answers.

Artificial Intelligence and Machine Learning (ML)  are over-hyped with very high expectations from various stake holders. Many organizations have embarked on this journey and are in various stages of implementation. In this session we will discuss Pfizer's AI/ML journey in Clinical Development, a few successful outcomes, best practices, and the successful use of NLP/ML in Pfizer's portfolio.

Being able to predict whether an individual is more susceptible to Adverse Drug Reactions (ADRs) is incredibly useful in both research and clinical context. Pharmacogenomics companies can leverage the UK Biobank, a large-scale biomedical database, to better understand how genes affect a person’s response to drugs. Learn how DNAnexus Apollo efficiently analyzes this massive dataset with explainable machine learning models to gain insight into ADRs.

We introduce Pangaea's Intelligence Extraction and Intelligence Extraction and Summarization (PIES), a neural architecture, which uses a unique data augmentation procedure to address data sparsity (training with only 200 examples). Coupled with the copy mechanism, PIES ensures model interpretability and precision of values, which appear in the output (textual narratives). PIES is generalizable for various input datasets. The outputs are validated by human experts which allows efficient model improvement with minimal human effort.

Pangaea's Intelligence Extraction and Summarization (PIES) incorporates novel unsupervised NLP and NLG demonstrating significantly higher accuracy and real world application against generic language models like GPT-3. PIES has effectively helped the pharmaceutical industry to: discover new clinical features to characterize hard to diagnose conditions and find more patients, especially undiagnosed and misdiagnosed; auto-generate textual narratives for regulatory reports; extract and format adverse events; summarize patient records; generate synthetic data.

Development of computer-aided de novo design methods to discover novel compounds in a speedy manner to treat human diseases has been of interest to drug discovery scientists for the past three decades. In the beginning, the efforts were mostly concentrated to generate molecules that fit the active site of the target protein by sequential building of a molecule atom-by-atom and/or group-by-group while exploring all possible conformations to optimize binding interaction with the target protein. In recent years, deep learning approaches are applied to generate molecules that are iteratively optimized against a binding hypothesis (to optimize potency) and predictive models of drug-likeness (to optimize properties). Synthesizability of molecules generated by these de novo methods remains a challenge. This review will focus on the recent development of synthetic planning methods that are suitable for enhancing synthesizability of molecules designed by de novo methods.

Biomedical knowledge graphs have exploded in popularity in past years, providing a framework for integrating and querying insights which conventionally have been siloed.  What was once novelty has evolved into a powerful commodity for accelerating drug discovery by enabling unprecedented views of biology. Janssen has conceptualized knowledge graphs beyond the convention of standalone network visualizations and rather embedded them within research workflows, facilitating adoption across its therapeutic areas. Recent use-cases will be highlighted to demonstrate the platform’s (1) extensibility in performing diverse analyses through its API (2) integration with Jupyter Notebooks (3) coupling with raw biology to surface novel targets.

Data is critical for biopharma organizations, but this data often comes in the form of language – unstructured and varied across sources – making it difficult to process and gain insights at scale. This session will dive into how you can automate human-like understanding of language data, allowing for large quantities of scientific literature, clinical studies, and medical reports to be analyzed.  Reduce the hours spent reading documentation to empower your organization with accelerated drug discovery and development efforts and cutting-edge medical innovations.

GPUs and AI/ML are used in multiple markets to accelerate results, or completely enable workflows that wouldn’t have been possible. At the core of these use cases is a requirement for massive amounts of data, and the need to enable researchers with the agility needed to conduct experiments and research. This session will review a use case of AI/ML for drug discovery and how its enabled by the WekaIO data platform.

This talk will summarize Pfizer's Knowledge Graph journey and highlight their lessons learned. By leveraging Knowledge Graph technology to make R&D data FAIR, Pfizer can surface actionable and meaningful insights and make knowledge accessible to and consumable by domain users. Pfizer's KG implementation helps accelerate R&D data discovery and understanding. It supports data scientists, researchers, scientists, and clinical and project leads in bringing medicine to the market faster by enabling advanced analytics on demand.

By automating HCS image analysis across Biopharma R&D, the Bio-IT award-winning software Genedata Imagence enables scientists to apply deep learning methods to simple and complex phenotypic imaging assays. Intuitive and interactive data representations enable scientists to train neural networks within minutes for fast, automated analysis of production screens. The scalable enterprise architecture of Genedata Imagence can use on-premise compute and cloud resources to serve a global organization.

There is a large amount of data generated outside of clinical trials. Real World Evidence (RWE), the analytical insights from Real World Data (RWD), plays an increasingly important role for Life Science companies both internally and with external stakeholders. However, data silos and interoperability challenges can limit the usability of data. With the right technology and ML/AI analytics infrastructure, different stakeholders can generate powerful RWE insights and visualizations.

Health and Life Sciences organizations’ storage systems, infrastructure and applications have grown organically as new data sources and associated workflows were developed to meet demands of innovative projects like genomic sequencing, drug discovery, and Cryo-EM.

This session will discuss:

- Notes from the field: Overcoming legacy environments

- Leveraging flexibility in hybrid-cloud drug discovery

- Simplifying data ownership at any scale whether on-prem, hybrid or cloud-native

Advancements in AI technologies, such as machine learning and deep learning, have provided new strategies to analyze large, multidimensional real-world data (RWD). I will provide an overview of the drug development studies that use both AI and RWD based on a review of articles from the past 20 years. I will also discuss current research gaps and future opportunities.

As infectious diseases such as COVID19 are raging the world, the demand for new vaccines is all time high. Current vaccine candidate identification approaches are ill suited to find vaccine candidates at whole proteome level. Vaxi-DL combines the strength of deep learning systems, immunoinformatics and text mining algorithms to achieve a combination of high speed, sensitivity and accuracy to predict vaccine candidates. Antigen identification is an important step in the vaccine development process. Here we present Vaxi-DL (https://vac.kamalrawal.in/vaxidl/), a web-based deep learning (DL) software that evaluates the potential of individual protein sequences as vaccine target antigens. It is designed to predict vaccine candidates for bacteria, protozoa, fungi, and viruses that cause infectious diseases in humans. The average validation accuracy obtained from the five iterations of the bacterial, protozoan, fungal, and viral models are 93%, 96%, 95%, and 92% respectively.