Cacher is the code snippet organizer for pro developers

# Metabolic Causal Planner A compact and interpretable deep learning framework for predicting cancer immunotherapy response from pre-treatment bulk RNA-seq profiles by integrating computational immunology, metabolic representation learning, and uncertainty-aware prediction. ## Overview **Metabolic Causal Planner** is designed for supervised patient-level cancer immunotherapy response prediction. Given a pre-treatment tumor bulk RNA-seq expression profile, the model predicts whether a patient is likely to respond to immune checkpoint blockade therapy. The framework integrates three coordinated modules: 1. **Counterfactual Metabolic Adjuster** Learns a latent biological state from transcriptomic profiles and estimates response-relevant metabolic perturbations in the latent space. 2. **Agent-driven Immune Response Mapper** Maps the adjusted molecular representation into an immune-response embedding to capture immune–metabolic interactions. 3. **Uncertainty-guided Response Predictor** Performs stochastic response prediction and estimates patient-level predictive uncertainty for robust decision making. An additional **Uncertainty-aware Refinement** strategy is used to regularize the final decision representation and reduce unstable or over-confident predictions. ## Key Features - Prediction from pre-treatment tumor bulk RNA-seq profiles - Immune–metabolic representation learning - Counterfactual latent metabolic adjustment - Stochastic uncertainty-aware response prediction - Entropy-guided refinement for robust classification - Internal evaluation on IMvigor210 - External validation on Hugo/GSE78220 - Comparison with classical machine learning, gradient boosting, neural, and immunotherapy-specific baselines ## Task The task is formulated as binary classification: - **Responder**: complete response or partial response - **Non-responder**: stable disease or progressive disease Given a normalized gene-expression vector, the model outputs a responder probability between 0 and 1. ## Datasets This project uses two public immune checkpoint blockade cohorts: | Dataset | Cancer Type | Treatment | Role | |---|---|---|---| | IMvigor210 | Metastatic urothelial carcinoma | Atezolizumab / anti-PD-L1 | Development cohort | | Hugo / GSE78220 | Metastatic melanoma | Anti-PD-1 | External test cohort | The IMvigor210 cohort is used for training, validation, and internal testing. Hugo/GSE78220 is used only for independent external validation. ## Model Architecture The overall framework follows the pipeline below: ```text Pre-treatment bulk RNA-seq | v Counterfactual Metabolic Adjuster | v Adjusted latent biological state | v Agent-driven Immune Response Mapper | v Immune-response embedding | v Uncertainty-guided Response Predictor | v Responder probability + predictive uncertainty Method Counterfactual Metabolic Adjuster The input RNA-seq vector is encoded into a compact latent biological state. A perturbation network estimates a counterfactual adjustment vector, which is added to the latent state to obtain an adjusted immune–metabolic representation. Agent-driven Immune Response Mapper The adjusted representation is transformed into an immune-response embedding using a neural mapping module with nonlinear activation, layer normalization, and dropout. Uncertainty-guided Response Predictor The predictor performs multiple stochastic forward passes using Monte Carlo dropout. The final responder probability is computed as the average prediction, while predictive uncertainty is estimated from the variance across stochastic predictions. Uncertainty-aware Refinement The model applies entropy-guided regularization and gradient-based feedback updates to refine the response decision representation before final classification. Evaluation Metrics The model is evaluated using: AUROC AUPRC Balanced Accuracy Macro-F1 Number of trainable parameters FLOPs per patient-level inference Confidence intervals, calibration analysis, Brier score, expected calibration error, and decision curve analysis may also be used for clinical AI reliability evaluation. Baselines The proposed method is compared with: Logistic Regression Random Forest XGBoost Multi-Layer Perceptron TIDE LightGBM All baselines use the same preprocessed RNA-seq input, label definition, data split protocol, and evaluation metrics. Installation git clone https://github.com/your-username/Metabolic-Causal-Planner.git cd Metabolic-Causal-Planner conda create -n mcp python=3.10 conda activate mcp pip install -r requirements.txt Requirements Recommended environment: Python 3.10 PyTorch 2.1.0 CUDA 11.8 NumPy 1.24.4 pandas 2.0.3 scikit-learn 1.3.2 SciPy 1.10.1 LightGBM 4.1.0 XGBoost Repository Structure Metabolic-Causal-Planner/ ├── configs/ │ └── default.yaml ├── data/ │ ├── raw/ │ └── processed/ ├── src/ │ ├── preprocessing.py │ ├── model.py │ ├── train.py │ ├── evaluate.py │ └── utils.py ├── scripts/ │ ├── run_train.sh │ ├── run_eval.sh │ └── run_ablation.sh ├── results/ ├── checkpoints/ ├── requirements.txt └── README.md Usage Data Preprocessing python src/preprocessing.py \ --config configs/default.yaml Training python src/train.py \ --config configs/default.yaml \ --seed 13 Evaluation python src/evaluate.py \ --config configs/default.yaml \ --checkpoint checkpoints/best_model_seed13.pt Repeated Runs for seed in 13 21 34 55 89 do python src/train.py --config configs/default.yaml --seed $seed python src/evaluate.py --config configs/default.yaml --seed $seed done Reproducibility The experimental protocol uses five random seeds: 13, 21, 34, 55, 89 For each run, IMvigor210 is split at the patient level into training, validation, and internal test subsets using an 8:1:1 stratified ratio. Hugo/GSE78220 is reserved as an independent external test cohort and is not used for training, validation, normalization fitting, hyperparameter tuning, or early stopping. Data Availability The raw datasets are obtained from public immunotherapy cohort resources. Due to data-access requirements, raw patient-level data are not redistributed in this repository. Scripts are provided to reproduce the processed matrices from the original public sources, subject to the corresponding data-access terms. Model Availability Trained model checkpoints, prediction files, split identifiers, and evaluation outputs can be deposited in an archival repository such as Zenodo after release. Citation If you use this repository or find this work useful, please cite: @article{fang2025metaboliccausalplanner, title={Integrating Computational Immunology and Metabolic Profiling for Cancer Immunotherapy Response Prediction from Pre-treatment Bulk RNA-seq Profiles}, author={Fang, Xuan}, journal={}, year={2025} } License This project is released for academic and research use. Please check the repository license file for details.

You are given an integer array nums and an integer target. Return the number of subarrays of nums in which target is the majority element. The majority element of a subarray is the element that appears strictly more than half of the times in that subarray.

You are given three integers n, l, and r. A ZigZag array of length n is defined as follows: Each element lies in the range [l, r]. No two adjacent elements are equal. No three consecutive elements form a strictly increasing or strictly decreasing sequence. Return the total number of valid ZigZag arrays. Since the answer may be large, return it modulo 109 + 7. A sequence is said to be strictly increasing if each element is strictly greater than its previous one (if exists). A sequence is said to be strictly decreasing if each element is strictly smaller than its previous one (if exists).

Locale settings are sometime missing in a brand new image; this is the fix

If you want the math behind it you can check wikipedia here https://en.wikipedia.org/wiki/Weibull_distribution Basically you have two parameters -the shape parameter (k), it represents the slope of the line, if it's exponential or logarithmic * k<1: decreasing death rate (infant mortality), early deaths are more likely * k = 1 : constant death (so same as the default life from Houdini) * k>1 : increasing death rate, death becomes more likely over time -the scale parameter (b), determines the location of the distribution, influences how quickly the death rate increases or decreases (e.g. a larger scale would shift distribution to the right, so longer life)

You are given three integers n, l, and r. A ZigZag array of length n is defined as follows: Each element lies in the range [l, r]. No two adjacent elements are equal. No three consecutive elements form a strictly increasing or strictly decreasing sequence. Return the total number of valid ZigZag arrays. Since the answer may be large, return it modulo 109 + 7. A sequence is said to be strictly increasing if each element is strictly greater than its previous one (if exists). A sequence is said to be strictly decreasing if each element is strictly smaller than its previous one (if exists).

Given a string text, you want to use the characters of text to form as many instances of the word "balloon" as possible. You can use each character in text at most once. Return the maximum number of instances that can be formed.

It is a sweltering summer day, and a boy wants to buy some ice cream bars. At the store, there are n ice cream bars. You are given an array costs of length n, where costs[i] is the price of the ith ice cream bar in coins. The boy initially has coins coins to spend, and he wants to buy as many ice cream bars as possible. Note: The boy can buy the ice cream bars in any order. Return the maximum number of ice cream bars the boy can buy with coins coins. You must solve the problem by counting sort.