With the increasing amount of biological data available and the advancement of analytical tools devoted to the extraction of meaningful information from such data we’re acknowledging a fundamental shift in life sciences. A lot of analytical efforts in biology are being transferred from wet lab workbenches to data centers and the involvement of bioinformaticians and data scientists in life sciences is increasing at an unprecedented rate. This process is particularly visible in cancer research were available databases (i.e. TCGA, EGA, dbGAP, etc.) allows practitioners to advance the understanding and control of cancer, predicting the development, progression, and malignancy of the disease in both populations and individuals. According to the organizers of the 30th Anniversary AACR Special Conference Convergence: Artificial Intelligence, Big Data, and Prediction in Cancer it is this interface of cancer biology, computation, and clinical oncology that will most likely produce the future breakthroughs in control of cancer. Ardigen’s Lead Scientist in Immunology, Giovanni Mazzocco was in Newport (RI) for this event. He attended presentations, talked to speakers, poster presenters and exhibitors. The conference was focused on discussing power and limitations of computational methods (i.e. machine learning, image processing, deep learning, etc.) in the analysis of cancer related data of different origins, a theme residing at the very heart of Ardigen’s vision for a truly personalized cancer treatment.
The Nobel prize-winner and MIT professor Phillip A. Sharp defined three big revolutions in life sciences: the first represented by advances in molecular and cellular biology (1953-1973), the second driven by the advent of genomic sequencing (2001-2009), and the third one (2009-current) characterized by the convergence of difference research fields (e.g. engineering, mathematics, physics, cancer biology) and technologies as a blueprint for innovation in life sciences. Particular emphasis was given to machine learning as a mean for extracting knowledge from complex multidimensional data. The president of Canadian Institute for Advanced Research Dr. Alan Bernstein, summarized the evolution of machine learning and artificial intelligence showing the potential of such methods especially in the context of cancer research, towards what he called the fourth revolution in cancer research.
Several speakers including Prof. Aviv Regev (MIT), Dr. Eliezer Van Allen (Dana-Faber), Dr. Anna Goldberg (SickKids, University of Toronto) and Prof. Andrea Califano (Columbia University), Dr. William C. Hahn (Dana-Farber), Dr. Pasquale Laise (Columbia University) and Prof. Alison P. Klein (Johns Hopkins), underlined the complex multi-factorial nature of cancer, were factors such as tumor heterogeneity, dynamic mutational landscape, resistance mechanisms, cancer–immune cross-talking and gene regulatory networks need to be taken into account for a faithful characterization of this disease, leading to better diagnostics, optimized treatment strategies, and discovery of novel therapeutic targets. Analytical strategies accounting for some of the mentioned aspect were proposed, including graph-theory models, Bayesian graphical models, interpretable deep neural networks and algorithms for the detection of master gene regulators.
Constructive criticism was directed towards the usage of machine learning methods were the biological interpretation of computational results is sometimes difficult or impractical (e.g. deep neural networks). This topic was raised by several speakers and attracted the attention of the audience at any given occasion. Models embedding biologically explicable causal-effect relations (e.g. Van Allen, Goldberg, Califano) were generally more appreciated and less criticized by the audience with respect to “black box” models with obscure biological interpretability. It was pointed out how the former models can be more useful than the latter ones in testing new scientific hypotheses and provide explanations for clinical effects. This trend was confirmed by discussions during networking events.
As mentioned by dr. Bernstein despite the abundance of Big Data in some area of cancer research (histopathology), it isn’t the case for all cancer data. In particular genomic data associated with clinical studies typically have a large number of features (e.g. gene mutations, gene expression) and a small number of observation (e.g. number of treated patients). Some modeling techniques can be useful in such cases. For example Dr. Goldberg showed the advantage of using a Bayesian approaches for the detection of relevant genes of interest in data-scarce situations.
The presented advancement of liquid biopsy, analytical techniques for the detection of circulating tumor Nucleic Acids (ctNAs) promotes the hope of non-invasive diagnostics for cancer detection and monitoring. Dr. Richard Klausner (founder and director of Juno Therapeutics, founder and director of GRAIL, chairman of Mindstrong and a director of AnchorDx), illustrated the three mandatory ingredients for the construction of highly specific (>99% specificity) ctNA platforms: high intensity sequencing (x60k sequencing depth) to detect cfNAs, large-scale clinical studies, a powerful machine learning approach. He also highlighted some issues related to cfNTs including the problem of Clonal Hematopoiesis of Indeterminate Potential (CHIP) and the large amount of material needed for sufficient cfNTs detection. Firm contrasting opinions about the practical usability of ctNA technologies were raised by Dr. Imran S. Haque (ex-CSO of Freenome) who severely doubted the maturity of such technology, sharing his experience in this field.
The importance of the interaction between cancer and immune system was the topic of several talks. Dr. Jerome Galon (INSERM Laboratory Integrative Cancer Immunology, founder of HalioDx) showed the immune contexture, which is determined by the density, composition, functional state and organization of the leukocyte infiltrate of the tumour, can yield information that is relevant to prognosis, prediction of a treatment response and various other pharmacodynamic parameters. Dr. Gerome shared the results of his decennial research on Immunoscore® an assay originally designed to predict the risk of relapse in early stage colon-cancer patients which was further shown capable of predicting metastases recurrence, extending the prognostic value of the immune tumor microenvironment in advanced-stage cancers (Cell 2018). Dr. Elizabeth M. Jaffee (Johns Hopkins) focused on neoepitope section for cancer vaccines and combination therapy based on cancer vaccine in association with checkpoints inhibitors and immune adjuvants (e.g. STING, OX40 agonists, etc).
Some sessions were dedicated to the predicting of cancer phenotypes through histopathological images leveraging both the large abundance of immunohistochemical data and powerful image processing tools such as Convolutional Neural Network (CNN).
Ardigen is glad to have attended the conference, gathering invaluable information instrumental to the development of its line of products for cancer immunotherapy.
The 30th Anniversary AACR Special Conference 2018 undoubtedly represented an historic milestone marking a new phase in cancer research, were clinicians, bioinformatics and data scientist works together to find novel solution in the fight against cancer.