Comparing the performance of statistical, machine learning, and deep learning algorithms to predict time-to-event: A simulation study for conversion to mild cognitive impairment.

Mild Cognitive Impairment (MCI) is a condition characterized by a decline in cognitive abilities, specifically in memory, language, and attention, that is beyond what is expected due to normal aging. Detection of MCI is crucial for providing appropriate interventions and slowing down the progression of dementia. There are several automated predictive algorithms for prediction using time-to-event data, but it is not clear which is best to predict the time to conversion to MCI. There is also confusion if algorithms with fewer training weights are less accurate. We compared three algorithms, from smaller to large numbers of training weights: a statistical predictive model (Cox proportional hazards model, CoxPH), a machine learning model (Random Survival Forest, RSF), and a deep learning model (DeepSurv). To compare the algorithms under different scenarios, we created a simulated dataset based on the Alzheimer NACC dataset. We found that the CoxPH model was among the best-performing models, in all simulated scenarios. In a larger sample size (n = 6,000), the deep learning algorithm (DeepSurv) exhibited comparable accuracy (73.1%) to the CoxPH model (73%). In the past, ignoring heterogeneity in the CoxPH model led to the conclusion that deep learning methods are superior. We found that when using the CoxPH model with heterogeneity, its accuracy is comparable to that of DeepSurv and RSF. Furthermore, when unobserved heterogeneity is present, such as missing features in the training, all three models showed a similar drop in accuracy. This simulation study suggests that in some applications an algorithm with a smaller number of training weights is not disadvantaged in terms of accuracy. Since algorithms with fewer weights are inherently easier to explain, this study can help artificial intelligence research develop a principled approach to comparing statistical, machine learning, and deep learning algorithms for time-to-event predictions.

The early evolutionary landscape of osteosarcoma provides clues for targeted treatment strategies.

Osteosarcomas are aggressive primary tumors of bone that are typically detected in locally advanced stages; however, which genetic mutations drive the cancer before its clinical detection remain unknown. To identify these events, we performed longitudinal genome-sequencing analysis of 12 patients with metastatic or refractory osteosarcoma. Phylogenetic and molecular clock analyses were carried out next to identify actionable mutations, and these were validated by integrating data from additional 153 osteosarcomas and pre-existing functional evidence from mouse PDX models. We found that the earliest and thus clinically most promising mutations affect the cell cycle G1 transition, which is guarded by cyclins D3, E1, and cyclin-dependent kinases 2, 4, and 6. Cell cycle G1 alterations originate no more than a year before the primary tumor is clinically detected and occur in >90% and 50% of patients of the discovery and validation cohorts, respectively. In comparison, other cancer driver mutations could be acquired at any evolutionary stage and often do not become pervasive. Consequently, our data support that the repertoire of actionable mutations present in every osteosarcoma cell is largely limited to cell cycle G1 mutations. Since they occur in mutually exclusive combinations favoring either CDK2 or CDK4/6 pathway activation, we propose a new genomically-based algorithm to direct patients to correct clinical trial options. © 2021 The Authors. The Journal of Pathology published by John Wiley & Sons, Ltd. on behalf of The Pathological Society of Great Britain and Ireland.

Germline RET variants underlie a subset of paediatric osteosarcoma.

BACKGROUND: Although considerable effort has been put into decoding of the osteosarcoma genome, very little is known about germline mutations that underlie this primary malignant tumour of bone. METHODS AND RESULTS: We followed here a coincidental finding in a multiple endocrine neoplasia family in which a 32-year-old patient carrying a germline pathogenic RET mutation developed an osteosarcoma 2 years after the resection of a medullary thyroid carcinoma. Sequencing analysis of additional 336 patients with osteosarcoma led to the identification of germline activating mutations in the RET proto-oncogene in three cases and somatic amplifications of the gene locus in five matched tumours (4%, n=5/124 tumours). Functional analysis of the pathogenic variants together with an integrative analysis of osteosarcoma genomes confirmed that the mutant RET proteins couple functional kinase activity to dysfunctional ligand binding. RET mutations further co-operated with alterations in TP53 and RB1, suggesting that osteosarcoma pathogenesis bears reminiscence to the stepwise model of medullary thyroid carcinoma. CONCLUSIONS: After Li-Fraumeni-predisposing mutations in TP53, RET becomes the second most mutated cancer-predisposing gene in the germline of patients with osteosarcoma. Hence, early identification of RET mutation carriers can help to identify at-risk family members and carry out preventive measures.

Convergent Evolution of Copy Number Alterations in Multi-Centric Hepatocellular Carcinoma.

In the recent years, new molecular methods have been proposed to discriminate multicentric hepatocellular carcinomas (HCC) from intrahepatic metastases. Some of these methods utilize sequencing data to assess similarities between cancer genomes, whilst other achieved the same results with transcriptome and methylome data. Here, we attempt to classify two HCC patients with multi-centric disease using the recall-rates of somatic mutations but find that difficult because their tumors share some chromosome-scale copy-number alterations (CNAs) but little-to-no single-nucleotide variants. To resolve the apparent conundrum, we apply a phasing strategy to test if those shared CNAs are identical by descent. Our findings suggest that the conflicting alterations occur on different homologous chromosomes, which argues for multi-centric origin of respective HCCs.

The evolutionary landscape of colorectal tumorigenesis.

The evolutionary events that cause colorectal adenomas (benign) to progress to carcinomas (malignant) remain largely undetermined. Using multi-region genome and exome sequencing of 24 benign and malignant colorectal tumours, we investigate the evolutionary fitness landscape occupied by these neoplasms. Unlike carcinomas, advanced adenomas frequently harbour sub-clonal driver mutations-considered to be functionally important in the carcinogenic process-that have not swept to fixation, and have relatively high genetic heterogeneity. Carcinomas are distinguished from adenomas by widespread aneusomies that are usually clonal and often accrue in a 'punctuated' fashion. We conclude that adenomas evolve across an undulating fitness landscape, whereas carcinomas occupy a sharper fitness peak, probably owing to stabilizing selection.

Exome sequencing of osteosarcoma reveals mutation signatures reminiscent of BRCA deficiency.

Osteosarcomas are aggressive bone tumours with a high degree of genetic heterogeneity, which has historically complicated driver gene discovery. Here we sequence exomes of 31 tumours and decipher their evolutionary landscape by inferring clonality of the individual mutation events. Exome findings are interpreted in the context of mutation and SNP array data from a replication set of 92 tumours. We identify 14 genes as the main drivers, of which some were formerly unknown in the context of osteosarcoma. None of the drivers is clearly responsible for the majority of tumours and even TP53 mutations are frequently mapped into subclones. However, >80% of osteosarcomas exhibit a specific combination of single-base substitutions, LOH, or large-scale genome instability signatures characteristic of BRCA1/2-deficient tumours. Our findings imply that multiple oncogenic pathways drive chromosomal instability during osteosarcoma evolution and result in the acquisition of BRCA-like traits, which could be therapeutically exploited.

Recurrent chromosomal gains and heterogeneous driver mutations characterise papillary renal cancer evolution.

Papillary renal cell carcinoma (pRCC) is an important subtype of kidney cancer with a problematic pathological classification and highly variable clinical behaviour. Here we sequence the genomes or exomes of 31 pRCCs, and in four tumours, multi-region sequencing is undertaken. We identify BAP1, SETD2, ARID2 and Nrf2 pathway genes (KEAP1, NHE2L2 and CUL3) as probable drivers, together with at least eight other possible drivers. However, only ~10% of tumours harbour detectable pathogenic changes in any one driver gene, and where present, the mutations are often predicted to be present within cancer sub-clones. We specifically detect parallel evolution of multiple SETD2 mutations within different sub-regions of the same tumour. By contrast, large copy number gains of chromosomes 7, 12, 16 and 17 are usually early, monoclonal changes in pRCC evolution. The predominance of large copy number variants as the major drivers for pRCC highlights an unusual mode of tumorigenesis that may challenge precision medicine approaches.

High-resolution breakpoint analysis provides evidence for the sequence-directed nature of genome rearrangements in hereditary disorders.

Although most of the pertinent data on the sequence-directed processes leading to genome rearrangements (GRs) have come from studies on somatic tissues, little is known about GRs in the germ line of patients with hereditary disorders. This study aims at identifying DNA motifs and higher order structures of genome architecture, which can result in losses and gains of genetic material in the germ line. We first identified candidate motifs by studying 112 pathogenic germ-line GRs in hereditary colorectal cancer patients, and subsequently created an algorithm, termed recombination type ratio, which correctly predicts the propensity of rearrangements with respect to homologous versus nonhomologous recombination events.