Data, AI, Infrustracture and Governance - SBEI 2023
Disclaimer:
I would like to acknowledge the Turrbal and Jagera peoples, Traditional Custodians of the land on which we gathered, and pay my respect to the their Elders past, present and extend that respect to all Aboriginal and Torres Strait Islander peoples.
Synopsis:
Symposium on Bioinformatics Engineering in Industry (SBEI) provides a dialog platform to include entrepreneurs, professionals and stakeholders from the life science, healthcare sector and industry partners under a same roof to exchange perspectives on major trending topics and challenges faced in bioinformatics field. This year, SBEI2023 was gathered in the picturesque Brisbane, the sunshinecapital
of Australia. The main themes for this year’s discussion were wrapping around bioinformatics as a service, AI and advocacy in governance policy.
Australia’s Bioinformatics Industry landscape
Australia has a vibrant bioinformatics presence that intersects with many fields including but not limited to healthcare, pharmaceuticals and farming. Although the bioinformatics industry is growing rapidly, small and medium firms, as well as start-ups are facing hefty headwinds post-pedantic. Established practises on the other hand are confronted with the growing threat of cybersecurity and dwefling capacity in afforable storage solutions, the lack of coordinated strategies in managing the explosion of genomics data sent uneasiness through our professional community.
Leverage analtyic AI in Bioinformatics context
AI has been a trendy topic in recent years, high enthusiasm has concentrated on the branch of regenerative AI. The bioinformatics community is bracing AI with open-arm, many attempts have been made in recent months to explore the potentials and the best areas in bioinformatics that AI has the capacity to made the most changes.
CSL, a Melbourne head-officed global biotech leader and is venturing into pharmaceuticals in recent years with the high profile acquisition of Swiss Vifor, demonstrated their passion and shared their strategic planning in regards to AI. CSL also highlighted their efforts in promoting innovation in AI space by establishing a self-sufficient data model and integrating private, public and proprietary data.
Microba, a medium-sized Brisbane-based microbiome company, has shared their efforts in exploring the potentials of AI in microbiome research. Microba has been working with the University of Queensland to develop AI models to predict the microbiome composition and the impact of microbiome on human health. Microba showcased their recent study in inflammatory bowel disease (IBD) and demonstrated the AI-enabled network analytics capacity in greatly improving the interpretability of microbiome interactome in humans.
Small business and start-ups also have a strong presence at this year’s SBEI. Dose/Me, a small Brisbane-based start-up showcased their product on AI-enabled optimum precision dosing for renal impairment patients. The product is distributed as a software plugin that can be integrated into the hospital’s electronic medical record (EMR) system, the systemm is currently under trial with a few hospitals in both Australia and the US. Dose/Me also shared their journey in development and commercialisation of the product, highlighted the funding challengings in post-pedantic era and emphesised the governance and complience gaps in data security. Dose/Me also demonstrated a pragmatic attitute towards AI to encourage the promotion of AI-enabled productivity growth rather than FOMO driven AI adoption, which I personally think was very refreshing. Life Whisperer is a Adelaide-based start-up and is contributing on AI-enabled embryo selection for IVF. Life Whisperer has been working with IVF clinics in Australia and the US to develop and commercialise the product. Life Whisperer also passionately shared their journey in AI-enabled image-based non-invasive embryo selction process and the challenges in commercialisation. Life Whisperer also shared their vision in the future of AI in IVF and the potential of AI in other areas of healthcare.
Data Availablility, Governance and Ethics
The rapid adoption of modern high-throughput sequencing and spectrometry technologies in clinical settings led to an unprecedented rate of health data generation. Health data is often sensitive due to its close association with patient welfare, with granular molecular information now being routinely captured in the light of personalised medicine, it creates a unique double-edged sword for medical practitioners and researchers. On one hand, the emergence of novel molecular technologies and abundant patient samples allow the cohort study of complex human pathophysiology that encourage cutting-edge medical discoveries to ultimately benefit all; on the other hand, the shear volume of modern digital health data poses a unique challenge in management, storage and security, increasing frequency of cyber misconducts and the severity of data leakage or data theft has been a prime concern in recent years, especially in the light of a few high profile instances nationally and globally. Furthermore, data governance and ethics have hence become more of an important topic to improve not only sensitive data security, but also streamline the compliance aspect of data generation and promote a data standard to facilitate insights generation.
In response to the unprecedented era of data Casanova, the pace of data generation has outstripped our current ability of insights generation by far. To mitigate the capability gap, consensus calls for a more efficient data utility. Data availability and accessibility were the two areas identified that need improvements. A robust discussion was erected in the context of the healthcare sector. Research, clinical and digital health are the trilogy stakeholders in the genomic data value chain, each stakeholder has a different set of priorities and sometimes even conflicting interests under the current genomics data landscape in Australia. The research community is advocating for open access to the data and emphasises the need to bridge data generation to ethics compliance and grant conditions; on the clinical front, the requirements for compliance are more prevalent; The digital health community on the other hand has a pronounced apitite on data security and integrity. Consequently, clinical and digital health communities lack motivation for an open-data approach given the associated efforts in identifying protection and elevated risks of cyber security/compliance breaches. Australian Genomics (AG) keynoted the discussion and highlighted the current genomics data management landscape confronted by a high degree of “data silo” and incompatibility between research and clinical genomics data management systems. AG further pointed out the lack of genomics data mobility from clinical to research, despite the historically high returns. AG also communicated a novel initiative to establish a national genomics data platform to promote genomics data interpretability and combat data segregation. GA also acknowledged the need for an overarching data standard for clinical genomics data, although there was currently no consensus pathway was identified. Noticeably, the demand for the democratization of genomics data is pronounced from the floor, which is believed to promote both research outputs and the high quality of evidence-based healthcare by improving transparency and knowledge sharing. The community has demonstrated a desire to further explore the best practices and standards in genomics data governance and sharing by engaging government authorities, industries and the public.
AI in Bioinformatics and Challenges
Under the renewed public enthusiasm towards a new generation of AI, the bioinformatics industry embraced the hype with open arms, and a few projects showcased the progress made with the adoption of deep neural networks. Google engineers demonstrated deepconsensus to address the common high error rate for PacBio circular consensus sequencing and deepvariant to achieve high accuracy and superior speed for variant calling on both PacBio and Illumina datasets (with benchmarked turnaround of less than 10 minutes for shotgun WES samples). CSIRO Transformational Bioinformatics Unit on the other hand showcased their efforts in building bioinformatics platforms and toolsets to integrate silo datasets to enhance collaboration and promote genomics to phenotype interpretations, serverless beacon initiative and BitEpi were introduced. VarientSpark was highlighted as a solution to the traditionally difficult task of associating high dimensional SNP markers to phenotypic traits as an example of explainable ML via random forest ML algorithm. Under the same theme, Life Whisperer gave the audience a different taste of modern healthcare with AI in the context of IVF applications. The team demonstrated a superior selection advantage in image-based aneuploid detection, which led to the introduction of a revolutionary continuous scaling system as opposed to the traditional categorical determination. This new capacity enables the treatment coupled with a degree of autonomy in the decision-making process which was not imaginable.
Despite the heightened community enthusiasm and successful demonstrates of AI potential, many challenges limit the AI uptake rate and impacts in bioinformatics community at a larger scale. In addition to persistent funding shortages, identifying appropriate design/application scopes that can produce an evidence-based demonstration of AI’s clear advantage is often difficult. The promotion of AI-based algorithms is also not without challenges, particularly in the lack of clear guidelines from regulatory authorities like the FDA and the lack of credited AI framework further exacerbated regulatory risks for AI in healthcare applications, this issue is even more pronounced in the context of generative AI. Simultaneously, the lack of early adopters and champion role models in general from the healthcare sector is not helping AI to gain more broad attractions either. The black-box nature of modern AI drew conflict sentiments arguably due to the lack of explainability, which is in contrast to the Platonism 1 thinking which modern scientific methods are based on. The panel discussion proposed a novel definition of explainability, which focuses on the instructions given to AI to generate reproducible outcomes rather than the underlying AI mechanism, which I think is rather eye-opening.
For future AI adoption in bioinformatics, the consortium reflects on the benefits of a potential uniform data structure for clinical genomics data to accelerate the training/development of AI models for genomics applications in the healthcare sector. The floor raised the growing concern in the community for the lack of coordination strategies in managing the explosion of genomics data. Australian Genomics, a national collaboration that supports genomics research to clinical practice transition, restated its commitment to a national genomics data-sharing platform to support the bioinformatics translation, which may provide some anxiety relief in this regard. The speaker panel also proposed a potential stacking strategy to allow different AI algorithms to self-correct the error and reinforce the positive signal hence producing a more accurate outcome. I can see maybe there is a potential application of such a strategy in morphology-based pathological diagnosis, nevertheless, I am sceptical about the accessibility of a multi-dimensional comprehensive expert-annotated training set, which may again echo the needs/benefits for clinically generated dataset to be made available for public access but of cause no compromise on patients’ privacy should ever be sacrificed.
Career topics and making an impact
At the end of the day program, the consortium derived a consensus that bioinformaticians still face challenges in lack of professional recognition and job security and career pathways; the drive for a routine recognition of bioinformatics tools to be routinely recognised can lead to a pathway for a better prospect of the profession.
Professional challenges
The lack of a clear definition of bioinformatician as a profession is one major contributing factor that underlines many challenges facing the profession, including the trap of constantly self-approving; lack of a clear career path; inconsistent job security and subdued public recognition. The status quo of bioinformatics broadly reassembles many trades including data scientists/engineers, statisticians, ecologists, molecular biologists/chemists, software engineers/architects, automation/orchestration engineers and many more. The loose definition of the trade not only creates identity confusion but also makes promotion and potential regulation of the profession difficult. The complex working environment that bioinformaticians are often subject to is not helping the profession to achieve its full recognition either. Bioinformaticians often work with a group of diverse but highly specialised professionals, in a healthcare setting for example, bioinformaticians routinely coordinate domain pathologists, a whole variety of medical specialists, molecular medical researchers and sometimes surgeons without mentioning a whole workforce of highly specialized instrument technicians and laboratory scientists. The nature of bioinformatic workflows often requires bioinformaticians to have unreserved exposure to the entire data value chain from data generation to data consumption. A constant presence at many steps along the data workflow often dilutes the professional image of a bioinformatician, which leaves the public to resemble an image about a bioinformatician to be like a ‘jack of all trades’. Under the current workplace reward framework, a well-defined, highly specialised workforce has an unparalleled advantage to be recognised and compensated properly.
Although there is a strong consensus on the need for a clearer definition of the profession, the cohort also recognises this overarching discussion will need to be referred to a different venue. Instead, this cohort devoted the efforts to focus on more immediate action plans that can bring immediate tangible improvements. Many topics were discussed including advocating for a defined career pathway in academia; potential regulations for clinical bioinformaticians and many more, one clear call for better recognition of bioinformatics tools drew a unanimous response.
Making impact with Bioinformatics tools and Bioinformatics as a Service (BaaS)
Bioinformatics software/tools are tangible assets bioinformaticians contribute towards industries, which can be positioned to elevate the recognition of the profession. The cohort reflected on the unparalleled subtle recognition received under the current reward system. To encourage more production of high-quality bioinformatics software/tools. A few proposals have drawn my attention including the introduction of a software standard to regulate software packaging, distribution and marketing to introduce a uniform experience and reduce the friction at the consumer level to allow better delivery of its intended functions. The potential of a uniform marketplace was also debated, although the idea dwindled in front of the classic dilemma between centralization and democratization.
A proposal for bioinformatics as a service seems to attract sizable attention. In the space of genomics in particular, the appetite for a service-based bioinformatics solution from industry partners has been growing, which may be attributed to the increasing complexity and cost of operating modern bioinformatics software. The bioinformatics solution as a service can leverage scale benefits to provide a competitive advantage in the market and, hence does have the potential to be a novel career path for well-positioned bioinformatican.
A summary and personal impression
SEBI2023 covered a broad range of topics that are directly relevant to the profession of bioinformatics with a good spread of both technical and career topics. The program captured the trendy topics of AI and showcased a variety of the potential applications of AI in bioinformatics, expanded discussion on infrastructure and open data access highlighted the current pain points in developing more AI-enabled bioinformatics applications. The confrontation by the lack of a common data structure standard reflects to some degree the composition of the bioinformatics community, which is by many vibrant but physically scattered groups. Conferences like SEBI2023 hence are even more important in creating an opportunity to form a united front before attacking more of a broad issue that requires more coordinated efforts. I found the career topics also drew incredible passion from the floor, the avenue to allow open discussion on many career topics seemed to be very well appreciated.
Moreover, the unique style of SEBI2023 brought industry, academia, entrepreneurs and supporting professionals under the same roof, which in turn created a unique flavour of discussions. The fully extended discussion with diverse perspectives sparked some unconventional but highly innovative ideas, which certainly is the part I enjoy the most. I also found the ‘melting-pot’ approach provides a more complete picture and helped my understanding of many complex concepts.
I would like to take this opportunity to express my gratitude towards the organising committee of SEBI2023 for this unique experience and acknowledge the great efforts of the committee to coordinate such a complex program. The small feedback if I may speak, I feel like an additional perspective from the governing regulatory bodies will help to complete many of the discussions we had. SEBI2023 was no doubt the most rounded conference experience I had in 2023, and for that, I am grateful for all the presenting speakers, participants, organising committee and all the friends old and new to make this experience complete.
Continue to ABACBS2023 …
References
-
Carré, M. H. (1955). Platonism And The Rise Of Science. Philosophy, 30(115), 333–343. doi:10.1017/S0031819100035002 ↩
Leave a comment