This work was carried out as part of a study of machine learning and artificial intelligence methods as applied to the field of biomedicine.
This work uses a dataset from the Breast Cancer Wisconsin (Diagnostic) Data Set. The aim of this work is to study and select the most suitable machine learning algorithms for identifying correlations between cell parameters and whether a sample is pathological or healthy. Tools: numpy, pandas, matplotlib, seaborn, scikit-learn Topics: correlation matrix, logistic regression, decision tree, random forest, gradient boosting, knn, cross validation, accuracy, F1 score, feature importances
Principal component analysis (PCA) transformation and visualization of the data from the article 'Genome-wide variation in the Angolan Namib Desert' by Sandra Oliveira et al. Tools: numpy, pandas, matplotlib, seaborn, scikit-learn Topics: PCA
PCA transformation, PCA and UMAP visualisation for bulk RNA-seq and scRNA-seq datasets. Tools: numpy, pandas, scanpy, matplotlib, seaborn Topics: PCA, UMAP, AnnData, log1p
Development a neural network model to predict pancreatic disease based on four urinary biomarkers (creatinine, LYVE1, REG1B and TFF1) and the blood plasma CA19-9 biomarker, which will be used for comparison with the urinary biomarkers. Tools: numpy, pandas, matplotlib, seaborn, scikit-learn, torch, tqdm Topics: heatmap, Iterative Imputer, MinMax Scaler, Logistic Regression, nn.Linear, SGD, Adam, BCEWithLogitsLoss, Weight Decay, Batch Norm, Dropout, ReLU, Tanh, Sigmoid, confusion matrix
The aim of this work is to practice working with the MOFA program in different modes for processing multi-omics data and to compare the results with PCA, UMAP, and several autoencoder algorithms. Tools: numpy, pandas, mofapy2, scikit-learn, torch Topics: MOFA, PCA, UMAP, Autoencoder
The aim of this project is to create a neural network model that can predict whether a person has an eye disease, using the Kaggle dataset. Tools: numpy, pandas, torch, torchvision, torchmetrics, timm, kornia, matplotlib, grad-cam, tqdm, seaborn, TensorBoard Topics: EfficientNetv2, augmentations, transfer learning, fine tuning, grad cam, cuda, autocast, TensorBoard
In this competition, participants are tasked with teaching their models to identify and segment FTUs in images from different human organs. The challenge lies in getting the models precise—accurately outlining each functional unit in every image. Tools: numpy, pandas, segmentation-models-pytorch, iterative-stratification, torch, cv2, tqdm, timm Topics: IoU, MultilabelStratifiedShuffleSplit, EfficientNet, augmentations, DeepLabV3Plus, DiceLoss Competition result: The final result of the competition: 1st place on the public dataset and 2nd place on the private dataset.