Covering Hierarchical Dirichlet Mixture Models on binary data to enhance genomic stratifications in onco-hematology.

Onco-hematological studies are increasingly adopting statistical mixture models to support the advancement of the genomically-driven classification systems for blood cancer. Targeting enhanced patients stratification based on the sole role of molecular biology attracted much interest and contributes to bring personalized medicine closer to reality. In onco-hematology, Hierarchical Dirichlet Mixture Models (HDMM) have become one of the preferred method to cluster the genomics data, that include the presence or absence of gene mutations and cytogenetics anomalies, into components. This work unfolds the standard workflow used in onco-hematology to improve patient stratification and proposes alternative approaches to characterize the components and to assign patient to them, as they are crucial tasks usually supported by a priori clinical knowledge. We propose (a) to compute the parameters of the multinomial components of the HDMM or (b) to estimate the parameters of the HDMM components as if they were Multivariate Fisher's Non-Central Hypergeometric (MFNCH) distributions. Then, our approach to perform patients assignments to the HDMM components is designed to essentially determine for each patient its most likely component. We show on simulated data that the patients assignment using the MFNCH-based approach can be superior, if not comparable, to using the multinomial-based approach. Lastly, we illustrate on real Acute Myeloid Leukemia data how the utilization of MFNCH-based approach emerges as a good trade-off between the rigorous multinomial-based characterization of the HDMM components and the common refinement of them based on a priori clinical knowledge.

Phase separation of ecDNA aggregates establishes in-trans contact domains boosting selective MYC regulatory interactions.

Extrachromosomal DNAs (ecDNAs) are found in the nucleus of an array of human cancer cells where they can form clusters that were associated to oncogene overexpression, as they carry genes and cis-regulatory elements. Yet, the mechanisms of aggregation and gene amplification beyond copy-number effects remain mostly unclear. Here, we investigate, at the single molecule level, MYC-harboring ecDNAs of COLO320-DM colorectal cancer cells by use of a minimal polymer model of the interactions of ecDNA BRD4 binding sites and BRD4 molecules. We find that BRD4 induces ecDNAs phase separation, resulting in the self-assembly of clusters whose predicted structure is validated against HiChIP data (Hung et al., 2021). Clusters establish in-trans associated contact domains (I-TADs) enriched, beyond copy number, in regulatory contacts among specific ecDNA regions, encompassing its PVT1-MYC fusions but not its other canonical MYC copy. That explains why the fusions originate most of ecDNA MYC transcripts (Hung et al., 2021), and shows that ecDNA clustering per se is important but not sufficient to amplify oncogene expression beyond copy-number, reconciling opposite views on the role of clusters (Hung et al., 2021; Zhu et al., 2021; Purshouse et al. 2022). Regulatory contacts become strongly enriched as soon as half a dozen ecDNAs aggregate, then saturate because of steric hindrance, highlighting that even cells with few ecDNAs can experience pathogenic MYC upregulations. To help drug design and therapeutic applications, with the model we dissect the effects of JQ1, a BET inhibitor. We find that JQ1 reverses ecDNA phase separation hence abolishing I-TADs and extra regulatory contacts, explaining how in COLO320-DM cells it reduces MYC transcription (Hung et al., 2021).

Random Walk Approximation for Stochastic Processes on Graphs.

We introduce the Random Walk Approximation (RWA), a new method to approximate the stationary solution of master equations describing stochastic processes taking place on graphs. Our approximation can be used for all processes governed by non-linear master equations without long-range interactions and with a conserved number of entities, which are typical in biological systems, such as gene regulatory or chemical reaction networks, where no exact solution exists. For linear systems, the RWA becomes the exact result obtained from the maximum entropy principle. The RWA allows having a simple analytical, even though approximated, form of the solution, which is global and easier to deal with than the standard System Size Expansion (SSE). Here, we give some theoretically sufficient conditions for the validity of the RWA and estimate the order of error calculated by the approximation with respect to the number of particles. We compare RWA with SSE for two examples, a toy model and the more realistic dual phosphorylation cycle, governed by the same underlying process. Both approximations are compared with the exact integration of the master equation, showing for the RWA good performances of the same order or better than the SSE, even in regions where sufficient conditions are not met.

Characterization and comparison of gene-centered human interactomes.

The complex web of macromolecular interactions occurring within cells-the interactome-is the backbone of an increasing number of studies, but a clear consensus on the exact structure of this network is still lacking. Different genome-scale maps of human interactome have been obtained through several experimental techniques and functional analyses. Moreover, these maps can be enriched through literature-mining approaches, and different combinations of various 'source' databases have been used in the literature. It is therefore unclear to which extent the various interactomes yield similar results when used in the context of interactome-based approaches in network biology. We compared a comprehensive list of human interactomes on the basis of topology, protein complexes, molecular pathways, pathway cross-talk and disease gene prediction. In a general context of relevant heterogeneity, our study provides a series of qualitative and quantitative parameters that describe the state of the art of human interactomes and guidelines for selecting interactomes in future applications.

Classification and Personalized Prognostic Assessment on the Basis of Clinical and Genomic Features in Myelodysplastic Syndromes.

PURPOSE: Recurrently mutated genes and chromosomal abnormalities have been identified in myelodysplastic syndromes (MDS). We aim to integrate these genomic features into disease classification and prognostication. METHODS: We retrospectively enrolled 2,043 patients. Using Bayesian networks and Dirichlet processes, we combined mutations in 47 genes with cytogenetic abnormalities to identify genetic associations and subgroups. Random-effects Cox proportional hazards multistate modeling was used for developing prognostic models. An independent validation on 318 cases was performed. RESULTS: We identify eight MDS groups (clusters) according to specific genomic features. In five groups, dominant genomic features include splicing gene mutations (SF3B1, SRSF2, and U2AF1) that occur early in disease history, determine specific phenotypes, and drive disease evolution. These groups display different prognosis (groups with SF3B1 mutations being associated with better survival). Specific co-mutation patterns account for clinical heterogeneity within SF3B1- and SRSF2-related MDS. MDS with complex karyotype and/or TP53 gene abnormalities and MDS with acute leukemia-like mutations show poorest prognosis. MDS with 5q deletion are clustered into two distinct groups according to the number of mutated genes and/or presence of TP53 mutations. By integrating 63 clinical and genomic variables, we define a novel prognostic model that generates personally tailored predictions of survival. The predicted and observed outcomes correlate well in internal cross-validation and in an independent external cohort. This model substantially improves predictive accuracy of currently available prognostic tools. We have created a Web portal that allows outcome predictions to be generated for user-defined constellations of genomic and clinical features. CONCLUSION: Genomic landscape in MDS reveals distinct subgroups associated with specific clinical features and discrete patterns of evolution, providing a proof of concept for next-generation disease classification and prognosis.

Development of machine learning models to prognosticate chronic shunt-dependent hydrocephalus after aneurysmal subarachnoid hemorrhage.

BACKGROUND: Shunt-dependent hydrocephalus significantly complicates subarachnoid hemorrhage (SAH), and reliable prognosis methods have been sought in recent years to reduce morbidity and costs associated with delayed treatment or neglected onset. Machine learning (ML) defines modern data analysis techniques allowing accurate subject-based risk stratifications. We aimed at developing and testing different ML models to predict shunt-dependent hydrocephalus after aneurysmal SAH. METHODS: We consulted electronic records of patients with aneurysmal SAH treated at our institution between January 2013 and March 2019. We selected variables for the models according to the results of the previous works on this topic. We trained and tested four ML algorithms on three datasets: one containing binary variables, one considering variables associated with shunt-dependency after an explorative analysis, and one including all variables. For each model, we calculated AUROC, specificity, sensitivity, accuracy, PPV, and also, on the validation set, the NPV and the Matthews correlation coefficient (ϕ). RESULTS: Three hundred eighty-six patients were included. Fifty patients (12.9%) developed shunt-dependency after a mean follow-up of 19.7 (± 12.6) months. Complete information was retrieved for 32 variables, used to train the models. The best models were selected based on the performances on the validation set and were achieved with a distributed random forest model considering 21 variables, with a ϕ = 0.59, AUC = 0.88; sensitivity and specificity of 0.73 (C.I.: 0.39-0.94) and 0.92 (C.I.: 0.84-0.97), respectively; PPV = 0.59 (0.38-0.77); and NPV = 0.96 (0.90-0.98). Accuracy was 0.90 (0.82-0.95). CONCLUSIONS: Machine learning prognostic models allow accurate predictions with a large number of variables and a more subject-oriented prognosis. We identified a single best distributed random forest model, with an excellent prognostic capacity (ϕ = 0.58), which could be especially helpful in identifying low-risk patients for shunt-dependency.