Starting to integrate AI for treatment decisions in oncology.
This study published in NPJ Precision Oncology describes a computational method to select the most effective targeted therapies for cancer patients based on the individual molecular profile of their tumor. Here we tell the story that led to the development of this new tool.
Precision oncology is a beautiful concept of individually treating cancer patients based on the molecular cause of their disease. However, we must overcome several theoretical and technological hurdles to make precision oncology a reality for all cancer patients.
The first hurdle was when we realized that most cancers are heterogeneous at the molecular level. In other words, a specific tumor type, for example, lung cancer, is driven by several different cancer genes. Therefore, it is not enough to know the tumor's localization and histology to use most targeted therapies successfully. Instead, we have to do a molecular diagnostic test on the tumor tissue to determine which driver gene is responsible for our patient's cancer.
For example, gefitinib, the inhibitor of the EGFR gene (Epidermal Growth Factor Receptor), failed in unselected lung cancer patients in 2003 (1). However, we identified a lung cancer patient whose tumor harbored an activating mutation in the EGFR gene in the same year (2). Our medical team treated her with gefitinib based on this molecular information. As a result, she achieved a complete dramatic response and remained in remission for over five years (2). Experiences like this case deepened our dedication to advance precision oncology (3-5).
It became apparent that we have to integrate the development of predictive biomarkers into drug discovery to co-develop and register molecularly targeted therapies and companion diagnostic tests, as we also advocated in our review in 2010 (6). For example, gefitinib was registered for lung cancer patients whose tumor harbors activating mutations in the EGFR gene, representing around 5% of lung cancer patients (6).
Meanwhile, cancer genome studies revealed more and more cancer-related "driver" genes and genetic alterations in these genes. Today, we know over 600 cancer genes and 6 million somatic mutations in cancer (7). The Pan-Cancer Analysis of Whole Genomes, published in 2020, revealed additional functional relevant genetic alterations in the non-coding region of the genome in one-third of patients. At least, now we know the whole picture. The study was able to identify a known driver genetic alteration in 95% of cases. The study also showed that each patient harbors a combination of an average of 4-5 alterations (8).
It is reassuring to know that even though the human genome is vast, it is a closed box. The number of potential cancer genes and genetic alterations is limited. We see that the number of new cancer genes discovered reaching a plateau.
The number of 600 cancer genes does not seem formidable, but the problem is that 600 is more than 100. This means that the average frequency of a cancer gene is less than 1%. EGFR is an overrepresented cancer gene in the longtail distribution of cancer genes. It was an exceptionally easy target to validate even in randomized clinical trials. Most cancer genes are far less frequent. Let's also consider the number of different specific alterations within these genes and the potential combination of genetic alterations. It is impossible to generate a high level of clinical evidence for most alterations' functional and clinical relevance and register all these biomarkers as companion diagnostics.
Even if the patient has a genetic alteration in a gene that constitutes a companion diagnostics of a particular targeted therapy, the functional relevance of the specific genetic alteration can be unknown or contradictory. For example, in the case of EGFR, the specific mutation can influence which EGFR inhibitor can be effective (9). In addition, the presence of other alterations in other genes in the same tumor can influence the therapeutic response to the targeted therapy, as we have recently shown in the case of a patient with ALK fusion (10). In these situations, we would need randomized studies for all combinations of alterations to decide which treatment we should prefer. Clinical trials are not feasible for such small groups. The paradox is that we want to give personalized therapy, but we want to use the principles of evidence-based medicine.
The solution that we propose in our study published NPJ Precision Oncology is to develop a standardized AI-based method to consider all that we know about the interactions between driver cancer genes, drugable targets, and targeted therapies. Next, implement this method into software as a medical device. The advantage of explainable computational systems is that they "make" standardized and reproducible decisions. Therefore, the medical device we use to make the decision for personalized, individual treatment decisions can be tested in randomized clinical trials according to the rules of evidence-based medicine. This also means that we can move away from the single biomarker paradigm to multiple parameters, aggregated data-based approach. We envision that AI-oncology algorithms will become the new companion diagnostics. SHIVA01 was the first randomized clinical trial to test the performance of precision oncology in comparison to chemotherapy (11). We had the privilege to work together with Professor Prof. Christophe Le Tourneau, the principal investigator of SHIVA01, and his team to analyze the performance of precision oncology again, this time powered by artificial intelligence (AI) and starting to integrate AI into precision oncology.
1. Johnson DH. Gefitinib (Iressa) trials in non-small cell lung cancer. Lung Cancer. 2003 Aug;41 Suppl 1:S23-8. doi: 10.1016/s0169-5002(03)00136-3. PMID: 12867059.
2. Schwab R, Pinter F, Moldavy J, Papay J, Strausz J, Kopper L, Keri G, Pap A, Petak I, Oreskovich K, Mangel L. Modern treatment of lung cancer: case 1. Amplification and mutation of the epidermal growth factor receptor in metastatic lung cancer with remission from gefitinib. J Clin Oncol. 2005 Oct 20;23(30):7736-8. doi: 10.1200/JCO.2005.02.4760. PMID: 16234532
3. Schwab R, Petak I, Kollar M, Pinter F, Varkondi E, Kohanka A, Barti-Juhasz H, Schönleber J, Brauswetter D, Kopper L, Urban L. Major partial response to crizotinib, a dual MET/ALK inhibitor, in a squamous cell lung (SCC) carcinoma patient with de novo c-MET amplification in the absence of ALK rearrangement. Lung Cancer. 2014 Jan;83(1):109-11. doi: 10.1016/j.lungcan.2013.10.006. Epub 2013 Oct 19. PMID: 24192513.
4. Kocsis J, Árokszállási A, András C, Balogh I, Béres E, Déri J, Peták I, Jánváry L, Horváth Z. Combined dabrafenib and trametinib treatment in a case of chemotherapy-refractory extrahepatic BRAF V600E mutant cholangiocarcinoma: dramatic clinical and radiological response with a confusing synchronic new liver lesion. J Gastrointest Oncol. 2017 Apr;8(2):E32-E38. doi: 10.21037/jgo.2017.01.06. PMID: 28480077; PMCID: PMC5401859
5. Brauswetter D, Gurbi B, Varga A, Várkondi E, Schwab R, Bánhegyi G, Fábián O, Kéri G, Vályi-Nagy I, Peták I. Molecular subtype specific efficacy of MEK inhibitors in pancreatic cancers. PLoS One. 2017 Sep 28;12(9):e0185687. doi: 10.1371/journal.pone.0185687. PMID: 28957417; PMCID: PMC5619833.
6. Peták I, Schwab R, Orfi L, Kopper L, Kéri G. Integrating molecular diagnostics into anticancer drug discovery. Nat Rev Drug Discov. 2010 Jul;9(7):523-35. doi: 10.1038/nrd3135. Epub 2010 Jun 7. PMID: 20531274.
7. Tate JG, Bamford S, Jubb HC, Sondka Z, Beare DM et al. COSMIC: the Catalogue Of Somatic Mutations In Cancer. Nucleic Acids Res. 2019 Jan 8;47(D1):D941-D947.
8. ICGC/TCGA Pan-Cancer Analysis of Whole Genomes Consortium. Pan-cancer analysis of whole genomes. Nature. 2020 Feb;578(7793):82-93. doi:10.1038/s41586-020-1969-6. Epub 2020 Feb 5.
9. Urbán L, Dóczi R, Vodicska B, Tihanyi D, Horváth M, Kormos D, Takács I, Pápai-Székely Z, Póka-Farkas Z, Várkondi E, Schwáb R, Hegedüs C, Vályi-Nagy I, Peták I. Major Clinical Response to Afatinib Monotherapy in Lung Adenocarcinoma Harboring EGFR Exon 20 Insertion Mutation. Clin Lung Cancer. 2021 Jan;22(1):e112-e115. doi: 10.1016/j.cllc.2020.09.005. Epub 2020 Sep 18. PMID: 33082101.
10. Urbán L, Dóczi R, Vodicska B, Kormos D, Tóth L, Takács I, Várkondi E, Tihanyi D, Lakatos D, Dirner A, Vályi-Nagy I, Peták I. Efficacy of Incremental Next-Generation ALK Inhibitor Treatment in Oncogene-Addicted, ALK-Positive, TP53-Mutant NSCLC. J Pers Med. 2020 Aug 28;10(3):107. doi: 10.3390/jpm10030107. PMID: 32872120; PMCID: PMC7563786.
11. Le Tourneau C, Delord JP, Gonçalves A, et al. Molecularly targeted therapy based on tumour molecular profiling versus conventional therapy for advanced cancer (SHIVA): a multicentre, open-label, proof-of-concept, randomised, controlled phase 2 trial. Lancet Oncol. 2015;16(13):1324-1334. doi:10.1016/S1470-2045(15)00188-6