Impressions

Artificial intelligence is a game-changer for life sciences research and patient care

For two days, international experts in artificial intelligence in the life sciences discussed at the industry symposium "AI in the Life Sciences." They had traveled to Bonn at the invitation of Martin Hofmann-Apitius and Holger Fröhlich, both professors at the Bonn-Aachen International Center for Information Technology (b-it) and heads of the Bioinformatics department at Fraunhofer SCAI. Nevertheless, it was not a scientific conference like any other: The symposium featured a balance of presentations by internationally well-known experts from academia and industry. This led to an exciting exchange about current research results in academia and R&D projects in the pharmaceutical industry. The participants very well received this unique conference format.

At the start of the symposium, Thomas Sattelberger, Former Parliamentary State Secretary from the German Federal Ministry of Education and Research, made it clear that he was no friend of excessive regulation in artificial intelligence. "I stand for experimentation and light-touch regulation," Sattelberger said. Sattelberger opposed the planned AI regulation by the European Parliament and the EU. He firmly believed this would weaken Europe and Germany's innovative power.

Mihaela van der Schaar of the Cambridge Centre for AI in Medicine at the University of Cambridge suggested that "all problems in medicine should be considered time-series problems." One goal of her research is to develop time-series models that can estimate the frequency of occurrence of an event, taking into account competing risks. In this way, predicting the probability of patients' conditions deteriorating during treatment is possible. For example, she has developed machine learning algorithms to estimate the progression of prostate cancer during active surveillance.

The last speaker of the first day, Bryn Roberts' talked about AI in healthcare. He emphasized that it includes: “detection” (e.g., digital biomarkers), “prediction” (risk models and prediction of clinical path), and “interaction,” if recommender systems and ChatBots take care of first-level healthcare support. But one of the main issues is the need for trust in medical AI. He predicted, "AI won't replace physicians, but physicians who use AI will replace those who don't. "

Intense discussions arose after Roland Eils' presentation on "Data save lives – deep learning from health data." Eils, founding director of the Center for Digital Health at the Berlin Institute of Health at Charité, showed how artificial intelligence could predict the course of diseases. For example, analyzing medical records and laboratory values can help generate fairly accurate percentages of whether one will develop certain disease conditions. Of course, such statements also have ethical implications, of which Eils is very aware. However, he expressed his belief, "If I could get a reliable statement that my behavior could reduce my risk of cardiovascular disease by 30 percent: I would change it."

Anna Bauer-Mehring reflected on the future empowerment of pharma research and development through AI. Her topics ranged from chemical spaces constructed and mined by AI via AI-driven optimization of therapeutic antibodies to the replacement of animal studies through in-silico models for ADME (Absorption, Distribution, Metabolism, and excretion) and Toxicity. Finally, she talked about using real-world data, e.g., from social media, to better understand patients' responses to drug treatment.

Day 1 – Introduction and welcome words

Prof. Dr. Martin Hofmann-Apitius

Welcome and setting the stage

© Fraunhofer SCAI
Prof. Dr. Martin Hofmann-Apitius

Prof. Dr. Martin Hofmann-Apitius gives the first warm welcome of the day, special greetings to Dr. Thomas Sattelberger and Uni Bonn, and thanks to the organizers.
 

To the talk

Prof. Dr. Jürgen Bajorath

Welcome and setting the stage

© Fraunhofer SCAI
Prof. Dr. Jürgen Bajorath

Prof. Dr. Jürgen Bajorath recapitulates the success story of the Life Science Informatics (LSI) curriculum at B-IT.

  • The key numbers:
    • international curriculum for 20 years
    • Bologna-conform before Bologna was implemented
    • all teaching in English
    • 80% of all Graduates (Master) continue with PhD studies, and approx. 90% of all graduates (Master & PhD) work in the industry later on
    • The demand for this “applied” curriculum is huge: we have 340 applications each year and accept only 25 students.
       

To the talk

Prof. Dr. Stefan Wrobel

Welcome by the B-IT Director

© Fraunhofer SCAI
Prof. Dr. Stefan Wrobel

Prof. Dr. Stefan Wrobel (Head of IAIS) made a strong point on the fact that “scientific excellence and impact on society are not mutually exclusive” and that the B-IT is actually bridging between science and industrial application. Furthermore, he highlighted the fact that the University of Bonn is amongst the top 100 worldwide.

To the talk

Prof. Dr. Dr. h. c. Michael Hoch

Welcome by the Rector of the University of Bonn

© Fraunhofer SCAI
Prof. Dr. Dr. h. c. Michael Hoch

Prof. Michael Hoch, Rector of the University of Bonn, highlighted that “B-IT has pioneered interdisciplinary research.” From his background as a Professor of Biology, he reflects on the ongoing paradigm change in the Life Sciences: “Computer science and AI are predicting systems behavior, and biologists are confirming the prediction.” 

He also stated clearly that “we need brilliant minds everywhere, not only in Universities” and that “B-IT has pioneered transfer activities in a Comprehensive University that is usually biased towards basic research.” 

To the talk

Dr. h. c. Thomas Sattelberger

Innovation and Freedom- Political Pragmatism in Innovation Policies

© Fraunhofer SCAI
Dr. h. c. Thomas Sattelberger

Dr. h. c. Thomas Sattelberger (former Parliamentary State Secretary, BMBF) reflected on the AI and innovation landscape in Germany, the EU and compared that to the situation in leading innovation cultures (UK and US). One of the key insights of his talk was the question of the balance between “Opportunities vs. Need for Regulation.” He outlined the principles of Ex Post and Ex Ante regulation: Germany, in particular, is characterized by anticipating the need to regulate the unforeseen (“Ex Ante”). Sattelberger sees a need to balance Ex Post and Ex Ante better and considers the need to constantly re-negotiate ethics and reaction (ex post) on technology. A “code of conduct” for new innovation will help us to deal with the required ex post regulation.

To the talk

Keynote 1: Machine Learning and revolutionizing healthcare

Prof. Dr. Mihaela van der Schaar

Transforming Healthcare: A Journey Through 5 Groundbreaking ML Innovations

Prof. Dr. Mihaela van der Schaar

Prof. Dr. Mihaela van der Schaar (Cambridge Centre of AI in Medicine University, University of Cambridge, UK) gave an overview of the opportunities of AI/ML for different applications in healthcare. Early diagnosis/risk models, specifically taking into account competing risks and different disease stages (a generalization of HMMs – Attentive State Space Models)

  • Irregular time series analysis, specifically bringing Neural ODEs into the Laplace domain (Neural Laplace)
  • Estimation of causal treatment effects (counterfactual recurrent network) using domain adversarial training. Irregular time series data are handled using control differential equations (TE-CDE).  Mihaela also points out a more recent further development of TE-CDE, which learns from the sampling intensity (TESAR-CDE).
  • AutoML (AutoPrognosis 2.0) for democratizing machine learning
  • Synthetic data 

Mihaela provides a vision of the application of AI/ML throughout the entire value chain of drug development.

To the talk

AI in Life Sciences: from academic research to industrial application

Dr. Norbert Furtmann

From Data to Predictions: Computational Optimization of Multi-specific Protein Therapeutics

© Fraunhofer SCAI
Dr. Norbert Fuhrmann

Dr. Norbert Furtmann (Sanofi Frankfurt) talked about the development of multi-specific protein therapeutics, which involves exploring vast design spaces that cannot be thoroughly explored through wet lab experiments alone. By utilizing valuable data resources generated using Sanofi's high-throughput protein engineering platform, AI-based virtual screening workflows were demonstrated to optimize the activity and expression profiles of multi-specific antibody variants.

To the talk

Dr. Raquel Rodríguez Perez

Can Machine Learning Replace ADME Experiments in Drug Discovery?

© Fraunhofer SCAI
Dr. Raquel Rodríguez Perez

Dr. Raquel Rodríguez Perez (Novatis Institutes for BioMedical Research) 

  • Machine learning (ML) predictions of ADME/PK compound properties are increasingly used in the pharmaceutical industry and are a valuable tool for compound prioritization even prior to synthesis
  • Large and diverse training sets (global) are preferred for ADME property predictions, whereas project-specific (local) models are prone to overfitting
  • In different studies, multi-task learning has helped to leverage in vitro data to predict an in vivo endpoint (e.g., brain penetration in rodents) as well as assay data for multiple species
  • Integrating model uncertainty in the predictions can help obtain ML predictions as accurately as experiments (on average). ML might assist in experiment selection if measurements focus on ‘inconclusive’ or ‘uncertain’model predictions

To the talk

Dr. Filip Miljković

Machine Learning Models for Predicting Human in Vivo PK Parameters Using Chemical Structure and Dose

© Fraunhofer SCAI
Dr. Filip Miljković

Dr. Filip Miljković (AstraZeneca) Filip has traveled from Sweden to talk about human pharmacokinetics. Human pharmacokinetics (PK) aims to solve the "right tissue" / "right safety" conundrum by studying the time course of ADME, including relationships to therapeutic and adverse effects of drugs. To this end, a large-scale data curation study of human PK clinical data has been attempted to build machine learning models for predicting human PK parameters using only existing chemical information and doses. Three fit-for-purpose models validated on internal AstraZeneca clinical candidates have been derived, which complement the ADME-derived PK predictions commonly used in practice.

To the talk

Dr. Paurush Praveen

From Cell Counting to Spatial Multiomics: The Role of ML and other Methods

© Fraunhofer SCAI
Dr. Paurush Praveen

Dr. Paurush Praveen (Miltenyi Biotec) talked about Miltenyi’s portfolio of AI/ML approaches in the fields of flow cytometry as well as multiplex microscopy-based imaging. 

 

To the talk

Dr. Ashar Ahmad

Causal Machine Learning in Drug Development

© Fraunhofer SCAI
Dr. Ashar Ahmad

Dr. Ashar Ahmad (Grünenthal Group) explained how causal machine learning models are leveraged at Grünenthal in the context of drug development, including ML-derived prognostic covariate adjustments, synthetic control arms, and assessment of comparative effectiveness of competitor drugs using real-world data. He closed by pointing out the greater vision of using causal ML throughout a value chain, starting with the analysis of real world and ending with ML-enhanced PK/PD models.

To the talk

Dr. Bryn Roberts

The Impact of Data Science and AI in Healthcare

© Fraunhofer SCAI
Dr. Bryn Roberts

Dr. Bryn Roberts (ROCHE) showed that healthcare systems are globally under stress (“in crisis” for a variety of reasons, including the aging population). He emphasizes the need to parallel the industrial revolution (“Industry 4.0”) in the healthcare system, and digitalization is key to achieving this goal. Precision Medicine, with bringing the right drug to the right patient at the right time, requires a much better understanding of disease etiology and pathophysiology mechanisms along with stratification of patients. 

AI in Healthcare includes “detection” (e.g., digital biomarkers), “prediction” (risk models and prediction of clinical path), and “interaction”, where recommender systems and ChatBots take care of first level healthcare support. He emphasized the need for trust in medical AI. 

He finished his talk with: “AI won’t replace physicians, but physicians who use AI will replace those who don’t”

To the talk

Day 2 – Keynote 2: AI in translational clinical research

Prof. Dr. Roland Eils

Data Save Lives – Deep Learning from Health Data

© Fraunhofer SCAI
Prof. Dr. Roland Eils

Prof. Dr. Roland Eils (Charité, BIH) recapitulated the AI „waves” that had an impact on biomedicine, with the 80s and 90s dominated by machine learning approaches and, more recently, deep learning taking over. He also emphasized on the role of Big Data in translational biomedicine and showed with the example of oncology how large-scale (interoperable) data can provide new insights into the genetic mechanisms underlying cancer. This knowledge is used for better translation, again by stratification of patients based on disease understanding and etiology. 

Large-scale studies across several indication areas showed that “age” and “sex” are amongst the strongest predictors of clinical outcomes for a variety of diseases; he also emphasized the utility of real-world data (such as EHRs) for risk assessment. Not surprisingly, the medical record of an individual is highly predictive of the personalized risk of that individual. Roland also emphasized the need to make clinical routine data better accessible for research and for individualized therapy.

To the talk

AI in medicine

Dr. Johann de Jong

Machine Learning for Genomics Driven Drug Discovery

© Fraunhofer SCAI
Dr. Johann de Jong

Dr. Johann de Jong (Boehringer Ingelheim) talked about Boehringer’s strategy for a more rational target prioritization (ideally ranking genes with multi-indication potential high) using machine learning. He pointed out the opportunities of using UK Biobank, FinnGen, and Genomics England data for this purpose. 

To the talk

Dr. Francisco Azuaje

Enabling Biomedical Research Through ML: From Exploration to Production

© Fraunhofer SCAI
Dr. Francisco Azuaje

Dr. Francisco Azuaje (Genomics England) gave a high-level overview of the ML strategy of Genomics England, comprising QC of genomes, LLMs for prioritization of genomic variants, de-identification of pathology reports as well as ML in the context of dedicated research use cases together with selected partners. Francisco dived into more details regarding the de-identification of pathology reports. In comparison to other models, the FLAIR method seems best suited.

To the talk

Dario Antweiler

Foundation Models for the Smart Hospital of the Future

Dario Antweiler

Dario Antweiler (IAIS) gave a guided tour through the Smart Hospital of the future. He highlighted the role of large language models (LLMs) for future interactions with patients and – like Roland Eils – emphasized the need for more digitalization in hospitals. One of the key application areas of AI in Smart Hospitals is focussing on reducing the administrative burden on medical professionals by, e.g., having AI write discharge letters. 

To the talk

Keynote 3: AI in personalized medicine

Dr. Anna Bauer-Mehren

Role of AI in Personalized Medicine

© Fraunhofer SCAI
Dr.Anna Bauer-Mehren

Dr. Anna Bauer-Mehren (ROCHE)  took us on a journey starting with FAIR data as a pre-requisite of useful big data that, in turn, paved the ground for applied AI in personalized medicine. Anna provided a sober reflection on the (future) empowerment of pharma R&D through AI; the guided tour took us from chemical spaces that are constructed and mined by AI via AI-driven optimization of therapeutic antibodies to the replacement of animal studies through in-silico models for ADME and Tox. 

Finally, she spotlighted the usage of real-world data, such as social media, for a better understanding of patients' responses to drug treatment.

To the talk

Applied AI

Dr. Shweta Bagewadi Kawalia

Transforming the Chemical Industry: Exploring BASF's Innovative Use of AI

© Fraunhofer SCAI
Dr. Shweta Bagewadi Kawalia

Dr. Shweta Bagewadi-Kawalia (Alumni, BASF) highlighted the role of BASF in everyday life: chemistry provided by BASF forms the basis of hundreds of products used by us on a day-to-day basis. Shweta focused on the required pragmatism in industrial AI. Sometimes, regular expressions do a perfect job and are much easier to implement than (re-)training a large language model for a specific domain.

Retrieval and classification of legacy documents from more than 100 years of chemical research at BASF is an objective where FAIR data, LLMs, and other AI services come together. Strongest impression with that talk: AI at BASF is solely driven by (internal) customers. No playgrounds.

To the talk

Stefan Geißler

Automatic Relation Extraction from Scientific Literature for Large Scale Knowledge Graph Creation

© Fraunhofer SCAI
Stefan Geißler

KAIRNTECH is a leading AI company in the field of text analytics. They use “AI on text” for information extraction, and Stefan Geißler showed an application use case that the company worked on together with Fraunhofer SCAI. 

The SHERPA workflow developed by KAIRNTECH identifies “cause-and-effect” statements in natural language (publications) and extracts these statements to build large knowledge graphs based on the syntax underlying the Biological Expression Language (BEL).  Stefan showed convincingly that entire indication areas can be represented as computable knowledge graphs covering major cause-and-effect relationships (e.g., pathophysiology mechanisms). 

To the talk

Dr. Philipp Senger

Towards a Prescriptive Field Testing Pipeline: How Data Science and Digital Technologies drive the Development of Future Crop Protection Products

© Fraunhofer SCAI
Dr. Philipp Senger

With the talk by Philipp Senger (Bayer AG) we “went green”. Patients were replaced by plants and clinical trials by field studies. Similarities between Crop Protection and Pharmaceutical Research were highlighted by Philipp, who provided a wonderful overview on digital farming, digital biomarkers in agriculture and crop protection and the application of AI methods in crop protection research. The “down-to-earth” approach making use of AI in a pragmatic approach that we learned about in the talk of Shweta Bagewadi-Kawalia (BASF) was echoed in Philipp Senger´s talk: pragmatic AI rules in industry. 

To the talk

Closing Remarks: Perspectives of Translational Data Science & AI

Prof. Dr. Holger Fröhlich

Closing Remarks: Perspectives of Translational Data Science & AI

© Fraunhofer SCAI
Prof. Dr. Holger Fröhlich

Prof. Dr. Holger Fröhlich concluded the symposium by saying that AI will be an integrated part of the life sciences. He emphasized the need to bring research together with industry and that the event demonstrated the breadth of the field of AI in life sciences and its interdisciplinary nature.

To the talk