Seminars 2023

16.01.2023 [online] Graph Neural Networks for Predicting DNA Secondary Structures

Speaker: Artyom Voytetsky, trainee researcher

The report presented three modules that carry out training, testing various graph networks based on the prepared data, and also draws additional graphs (ROC-, PR- and F1 score curves).

20.02.2023 [online] Z-DNA segment prediction pipeline

Zoom

Speaker: Anna Danilova, trainee researcher

The report has been updated by a program that is a pipeline for learning and runs a Transformer that predicts Z-DNA regions.

13.03.2023 [online] Inverted repeats in the eukaryotic genome: structure and functions

Zoom

Speaker: Aleksandr Fedorov, Junior Researcher

A significant, and often overwhelming, portion of eukaryotic genomes is composed of repetitive sequences. DNA repeats are extremely diverse in nature and often do not have a defined biological function. However, among the variety of eukaryotic repeats, inverted repeats (IRs) stand out separately. Based on a series of experiments, we know that some of them tend to form cruciform structures in DNA, which are involved in replication, transcriptional regulation, and chromatin organization. Additionally, during transcription, IR can form long segments of double-stranded RNA, which are critical for innate immune response and can further be processed into small interfering RNAs with various functions. These and other details of IR biology will be discussed during the presentation.

17.04.2023 [online] A non-cooperative framework for cooperative action: a generalized Nash program

Pokrovsky Boulevard 10, Auditorium R406

Speakers:  Dmitry Levando, Candidate of Economic Sciences

We demonstrate how to build coalition structures with any number of coalitions from non-coalitional actions of individual agents. The approach extends the traditional non-cooperative Nash game to the area of non-cooperative coalition formation. The proposed game includes a mechanism for forming a coalition structure and has two outcomes: the distribution of players among coalitions and payoffs for each player. An individual strategy is a tuple, a coalition structure and a strategy for it. The player has a set of strategies for each coalition structure. Coalition structures are described by Young diagrams. The (social) mechanism transforms the set of all individual strategies into a final one with an explicit distribution of players among coalitions. This mechanism has a coercive force to eliminate conflicts of individual choice. The final coalition structure has a set of individual payoffs. As usual, every game has a mixed strategy equilibrium, which is different from the results of cooperative game theory. The Nash game is a special case of the presented model. We demonstrate how to construct a noncooperative stability criterion for a strong Nash equilibrium.

17.05.2023 [online] Using Kas-seq to identify flipons

Zoom

Speakers: Konovalov Dmitry, trainee researcher

The report discusses the results of the analysis of the Kas-seq experiment performed for several human and mouse cell lines. This method allows, using ketoxal, to identify sections of single-stranded DNA. Such areas are formed and often signal the presence of flipons. Special attention was paid to areas in promoter regions. The results of computational experiments show that Kas-seq is consistent with other experimental data and predictions of machine learning models for identifying flipons.

21.06.2023 [online] Development of a pipeline for Z-RNA research

Zoom

Speakers: Danilova Anna, trainee researcher

When studying the role of Z-RNA in the functioning of the genome, it is necessary to analyze the secondary structure of RNA. Both the structure itself and the ZH-score value are of interest. Since it is necessary to consider various structures, manual search using ready-made solutions (for example, RNA-Fold) is time-consuming. In addition, existing packages do not allow calculating ZH-score. In this regard, there was a need to develop our own pipeline that would allow us to find various secondary structures for given input sequences, calculate the ZH-score of the resulting structures and visualize them. During the report, the work of the developed pipeline was demonstrated.

19.07.2023 [online] Plasma B cell-associated gene signature predicts better response to immunotherapy in metastatic but not primary melanoma tumors

Zoom

Speakers: Puzanov Grigory, researcher

In recent years, a number of studies have pointed to a role for B cells in predicting response to immunotherapy. However, this role varies depending on the tumor type. The report describes the signature of plasma B cells found through analysis of single-cell sequencing data from brain metastases for various tumor types (melanoma, lung cancer, breast cancer, ovarian cancer). For each sample in which B cells were present, a subtype of cells with specific expression of the ZBP1, DERL3 and TNFRSF17 genes was found. The TCGA data further confirms the association of the expression of these genes with plasma cells and reveals an association with better survival for melanoma metastases.

23.08.2023 [online] Mathematical modeling of the spatial structure of protein molecules

Zoom

Speakers: Ignatov Andrey Dmitrievich

Modeling the spatial structure of a protein is a computationally complex task associated with a large number of degrees of freedom, and therefore a large number of variables. To simplify this task, large-block models have been proposed to simplify the geometry of the protein molecule. The talk will discuss the large-block HP model based on the hydrophobicity and polarity of protein side chains. Using it, it is possible to reduce the problem of predicting protein conformation to a combinatorial optimization problem, in which the objective function is the number of contacts between H-monomers of the protein chain.

A number of techniques have been proposed to speed up the prediction of the spatial structure of a protein in the HP model. A new precise algorithm has been developed that makes it possible to create both maximally dense hydrophobic nuclei and nuclei with a limitation on the maximum number of contacts between H-monomers. In addition, methods for filtering hydrophobic cores and sets of constraints for placing a protein in a simulated hydrophobic core are proposed.

20.09.2023 [online] Application of machine learning to analyze treatment outcomes for coronary syndrome

Zoom

Speaker: Kirdeev Alexander, trainee researcher

The report presents a study aimed at analyzing the effectiveness of treatment for coronary syndrome using various data imputation methods. One of the main aspects of the work is the comparison of data imputation methods, such as IterativeImputer, Missforest, KNN, Mean imputation, in the context of handling missing data in medical datasets. The application and results of each method are reviewed to determine the optimal approach to handling missing values. Additionally, the work analyzes the quality of training using Optuna as an optimizer for selecting hyperparameters. This approach aims to improve model performance and prediction accuracy in the context of coronary syndrome management.

18.10.2023 [online] Biomarkers study to improve Machine Learning prediction of long-term risk of myocardial infarction, stroke, and cardiac death

Zoom

Speaker: Konstantin Burkin, trainee researcher

Currently, cardiovascular diseases occupy 1st place in the number of deaths per capita. A reduction in this indicator can be achieved through early diagnosis of patients. Improving such diagnostics is an important goal and is feasible through the use of biomarker data. The report shows comparison of a number of biomarkers and proves that the use of the PCSK9 biomarker allows a patient's risk to be assessed with high accuracy.

22.11.2023 [online] Study of biomarkers in the diagnosis and prognosis of treatment outcomes for coronary syndrome

Zoom

Speaker: Kirdeev Alexander, trainee researcher

The study provides an in-depth analysis of the importance of biomarkers, including NtProBnP and PCSK9, to identify their role in predicting treatment outcomes. The influence of the GRACE 2 integral scale on predicting and assessing the severity of coronary syndrome in the Russian population is also considered. The report not only highlights the importance of biomarkers in medical research, but also provides a comparative analysis of modern Automl solutions, highlighting the unique features of our development pipeline, tailored to the specifics of biomarker analysis in the context of coronary syndromes.