Goat_bones-ML

Goat_bones-ML predicts fetal skeletal development in goats by applying machine learning models to maternal bone metabolic protein biomarkers to quantify the impact of maternal nutrition restriction (MNR) during gestation.

Key Features:

• Data Integration: Utilizes 72 input datasets comprising one-hot encoding of experimental conditions, original serum metabolic data, experiment-centered features, and probabilities related to experimental conditions.
• Machine Learning Models: Implements seven machine learning methods to predict six fetal bone parameters: weight, length, and diameter of the femur and humerus.
• Prediction Targets and Performance: Predicts femur weight (Fw), femur length (Fl), femur diameter (Fd), humerus weight (Hw), humerus length (Hl), and humerus diameter (Hd), with SVM classifiers achieving leave-one-out cross-validation accuracies of 1.0 for Fw, Fl, and Hw; 0.98 for Fd; 0.946 for Hd; and 0.696 for Hl.
• Feature Engineering: Applies one-hot encoding of experimental conditions and computes moving averages of parathyroid hormone (PTH) within nutritional conditions as input features.
• Feature Importance Analysis: Performs feature importance analysis showing that experimental-conditions encoding combined with serum metabolic data is influential, with moving averages of PTH contributing variably to predictions depending on the bone parameter.
• Biomarker Inputs: Uses maternal serum bone metabolic proteins and biochemical markers including C-terminal telopeptides of collagen I (CTx), N-terminal telopeptides of collagen I (NTx), and bone alkaline phosphatase (BALP) as predictors.

Scientific Applications:

• Fetal development under MNR: Analyzes effects of maternal nutrition restriction during mid- and late-gestation on fetal bone parameters in goats using maternal serum bone metabolic proteins.
• Comparative mammalian skeletal research: Elucidates relationships between maternal bone metabolic biomarkers (including PTH, CTx, NTx, BALP) and fetal skeletal growth to support broader mammalian comparative studies.

Methodology:

Uses 72 input datasets with one-hot encoded experimental conditions, original metabolic data, experiment-centered features, and probabilities related to experimental conditions; computes moving averages of PTH; trains seven machine learning models including Support Vector Machines; evaluates performance with leave-one-out cross-validation; and performs feature importance analysis for the six fetal bone parameters.