sstefan01/methylation-survival ? reverse-engineered prompt

Build me a Python project that predicts survival for childhood brain tumor patients from DNA methylation beta values, copy number variation data, and metastasis status.

I want a simple workflow where I can train the model, run it on a held out cohort with a saved model, evaluate it with survival metrics like c index, time dependent AUC, and Brier score, and generate plots of the results. Please make it work from a single config file so I can point it at the data files and choose which cohorts to train and test on. The model should use a sparse first layer for the high dimensional methylation and CNV features, then a few normal dense layers, then a survival prediction head for discrete time survival output.

Please also include clear scripts for training, inference, evaluation, and plotting, and make sure the code can load pre trained weights if I do not want to retrain. If you need anything unclear, check the current docs online and make the defaults sensible.