BlastRNAPredict: Using Machine Learning to Predict IVF Success from RNA in Embryo Culture Medium

This article introduces the BlastRNAPredict project, which builds a machine learning model to predict pregnancy and live birth outcomes of IVF by analyzing RNA molecular features in blastocyst culture medium. Compared to traditional morphological assessment, this model has higher prediction accuracy and provides a new direction for non-invasive embryo evaluation. The project was developed by VafaeeLab and released on the GitHub platform on May 26, 2026.

In vitro fertilization (IVF) technology faces a core challenge: how to accurately predict embryo implantation and developmental potential. Currently, the widely used morphological scoring (observing embryo appearance under a microscope) in clinical practice is highly subjective and has limited accuracy. With the development of molecular biology and AI technology, researchers have begun to explore prediction approaches at the gene expression level, and BlastRNAPredict is a representative of this direction.

BlastRNAPredict is an end-to-end machine learning workflow with core steps including:

1. Data Preprocessing: Normalize RNA sequencing data using DESeq2;
2. Feature Selection: Use Bootstrap-enhanced LASSO method to screen gene subsets related to outcomes;
3. Model Construction: Adopt Ridge regression (main classifier) and random forest (comparative model);
4. Validation Strategy: Sibling-stratified 10-fold cross-validation (internal) and cross-center external validation to ensure model reliability.

The project dataset contains 490 samples, with 75 samples in the core BF cohort (blastocyst culture medium). Baseline model results show:

• Morphology-only model: Internal cross-validation AUC 0.561, external validation AUC 0.667, limited predictive ability;
• Age-only model: Internal AUC 0.626, external only 0.458, poor cross-center stability.

The value of BlastRNAPredict is reflected in:

• Scientific Aspect: Verifies the feasibility of using culture medium RNA as a biomarker for embryo quality;
• Clinical Aspect: Provides a non-invasive and objective embryo selection scheme, which is expected to improve the success rate of single embryo transfer and reduce the risk of multiple pregnancies;
• Methodological Aspect: The strict validation strategy (sibling-stratified cross-validation, cross-center validation) sets a benchmark for the field.

The project has limitations such as small sample size (only 75 samples in the core BF cohort), high cost of RNA sequencing, and unclear biological mechanisms. Future directions include: expanding sample size for multi-center validation, exploring combined RNA and morphological models, developing rapid detection methods, and integrating multi-omics data to improve accuracy.