It’s called Sphinks, sphinx. He is a brainy child of artificial intelligence who ‘speaks a little Italian’. In the sense that it was developed by an international team, coordinated by tricolor scientists in force in the USA. White coats struggling with difficult tumors such as glioblastoma, the most lethal brain cancer, will be able to question him to be guided towards possible more targeted, precision therapies. Ai’s new algorithm analyzes and combines all the identity cards of the neoplasm – the so-called ‘omics’ data, which photograph from genes to protein composition, from lipids to metabolites, and so on – and classifies the cancer according to all its characteristics, helping experts to identify potential therapeutic targets and target available weapons against them, as well as directing the search for new strategies and treatment options.
The study has earned the pages of ‘Nature Cancer’ and bears the signatures of a golden couple of Italian research: Antonio Iavarone, deputy director of the Sylvester Comprehensive Cancer Center of the Miller School of Medicine – University of Miami and research coordinator, and Anna Lasorella, co-coordinator of the work. The algorithm focuses on data from platforms analyzing tumor proteins and their modifications to identify enzymes, called kinases, that produce hallmarks in malignant cells. Specific inhibitors exist for many of these enzymes, making them potential therapeutic targets. In the research, the team developed and tested the Sphinks algorithm (which stands for Substrate Phosphosite based Inference for Network of KinaseS) first using data from the glioblastoma brain tumor. Then he extended the study to other human cancers as well.
Algoritmi e avatar – “The idea – explains Iavarone to time.news Salute – is to introduce for the first time tools that derive from basic research and computational medicine, through artificial intelligence and therefore machine learning, to analyze an enormous amount of data that today can be generated for each tumor and extract the most effective personalized therapies for each patient.We have worked to create a sophisticated system that allows us to be as sure as possible of the fact that that patient can respond to a specific inhibitor of a specific kinase. The other ally in this sense are the organoids, a very important component of our study: avatars that reflect the characteristics of the tumor and on which we test the drugs identified by the computational analyses. If they work, this gives us further assurance that they are can then have an effect on the patient”. The brainy ‘classifier’ will be “made available to any structure – even in Italy – who want to consult it to sort out the patients’ tumours”. The hope is that we will gradually “increase treatment and improve therapies for these complex tumours”.
“We will have a portal – continues Iavarone – and these centres, by inserting a whole series of necessary data, will be able to obtain the classification of the tumor in one of the groups that we have identified. At that point there is the possibility of introducing the patient into a different stratification therapeutics. Naturally, the drugs now have to be used through clinical trials that we hope to start carrying out progressively. It must also be said that some groups of tumors are easier to identify, the drugs are already available. For others, however, we must face reasoning with the pharmaceutical companies that have them to see if they can be used. And for still other groups, however, they really have to create them from scratch. What can be seen through these studies is that every single tumor can be different in terms of complexity of treatment”.
Sphinks identified Pkcd and Dnapkcs kinases as having the greatest impact on malignant features in two subtypes of glioblastoma. Using tumor organoids derived from patient tumors, the team demonstrated that drugs that interfere with the enzymatic activity of each of the kinases block tumor growth. These kinases are equally activated in the similar subtypes of breast and lung cancer. “Since its conception, the work had a translational objective, that is aimed at identifying therapeutic modalities different from those currently used for patients with glioblastoma”, highlights Iavarone. “Sphinks will offer clinical oncologists a new tool to apply selected treatments for specific subtypes of cancer. The correct classification of each tumor is the basis of this way of selecting personalized therapies.”
The mission is to bring a promise of better treatments to the bedside of patients who are today in complicated situations, such as those affected by glioblastoma, who face a poor prognosis: the 5-year survival rate is less than 10%. While numerous drugs are being developed to improve these numbers, physicians need tools that identify the fundamental molecular mechanisms that underpin each patient’s neoplastic disease in order to personalize care.
In previous work, Iavarone’s team had previously reported a classification of glioblastoma based on fundamental tumor cell processes that managed to group patients into four distinct groups based on specific tumor vulnerabilities. Lasorella explains that in this latest study the new classification method has been perfected “using artificial intelligence methods to integrate data from complex analyzes of genes, protein composition, lipids, metabolites and protein modifications that determine their activation or inactivation (acetylation and phosphorylation) applied to each tumor”.
Hunt for the Achilles heels of the most difficult tumors – The results obtained, Iavarone reiterates, demonstrate that the combination of new computational biology techniques with methods of comprehensive analysis of tumors (multi-omics) can generate knowledge for the introduction of therapies designed on the basis of the specific glioblastoma subclass for each patient. Although Sphinks was initially tested on glioblastoma, “the algorithm is equally applicable to many types of cancer,” the experts explain. “We are exploring the concept of ‘basket trial’ – says Iavarone – that is clinical studies that include patients with the same biological subtype in different tumors. If patients with glioblastoma or breast or lung cancer have similar molecular characteristics, they can be included in the same protocol clinician with the ability to quickly bring patients the most effective medicines possible for their cancers.”
Next steps? “We are now trying to understand the synergies between drugs to also define how the most important ones can be associated for each group”, concludes Iavarone.
How Sphinx works – Sphinks was first tested in glioblastoma, but the algorithm is equally applicable in many cancers. Leverage machine learning to refine omics datasets and create an interactome (a complete set of biological interactions) to focus on the kinases that generate aberrant growth and treatment resistance in each subtype of glioblastoma. These results show that multi-omics data can generate new algorithms that predict which specific targeted therapies may represent the best treatment options based on each patient’s glioblastoma subtype. “Now we can stratify patients with glioblastoma better and better,” says Iavarone. “Reading the genome alone wasn’t enough. They needed more comprehensive data to identify tumor vulnerabilities.”
The authors believe this approach could yield actionable information that could benefit up to 75% of glioblastoma patients. “This classifier can be used in virtually any laboratory,” explains Lasorella, professor of biochemistry and molecular biology and co-senior author of the study. “By importing omics data into the web portal, pathologists receive information on tumor classification.” Iavarone, Lasorella and colleagues believe that this information could eventually help create a new type of clinical trial.
