Optimized cell type signatures revealed from single-cell data by combining principal feature analysis, mutual information, and machine learning

[Exploration of the Predictive Value of Peripheral Blood-related Indicators for EGFR Mutations and Prognosis in Non-small Cell Lung Cancer Using Machine Learning]

BACKGROUND

Epidermal growth factor receptor (EGFR) sensitive mutation is one of the effective targets of targeted therapy for non-small cell lung cancer (NSCLC). However, due to the difficulty of obtaining some primary tissues and the economic factors in some underdeveloped areas, some patients cannot undergo traditional genetic testing. The aim of this study is to establish a machine learning (ML) model using non-invasive peripheral blood markers to explore the biomarkers closely related to EGFR mutation status in NSCLC and evaluate their potential prognostic value.

METHODS

2642 lung cancer patients who visited Jiangsu Cancer Hospital from November 2016 to May 2023 were retrospectively enrolled and finally 175 NSCLC patients with complete follow-up data were included in the study. The ML model was constructed based on peripheral blood indicators and divided into training set and test set according to the ratio of 8:2. Unsupervised learning algorithms were used for clustering blood features and mutual information method for feature selection, and an ensemble learning algorithm based on Shapley value was designed to calculate the contribution of each feature to the model prediction result. The receiver operating characteristic (ROC) curve was used to evaluate the predictive ability of the model.

RESULTS

Through the feature extraction and contribution analysis of the predictive results of the interpretable ML model based on the Shapley value, the top ten indicators with the highest contribution were: pathological type, phosphorus, eosinophils, monocyte count, activated partial thromboplastin time, potassium, total bilirubin, sodium, eosinophil percentage, and total cholesterol. The area under the curve (AUC) of the model was 0.80. In addition, patients with hyponatremia and squamous cell carcinoma group had a poor prognosis (P<0.05).

CONCLUSIONS

The interpretable model constructed in this study provides a new approach for the prediction of EGFR mutation status in NSCLC patients, which provides a scientific basis for the diagnosis and treatment of patients who cannot undergo genetic testing.

DataXflow: Synergizing data-driven modeling with best parameter fit and optimal control - An efficient data analysis for cancer research

Building data-driven models is an effective strategy for information extraction from empirical data. Adapting model parameters specifically to data with a best fitting approach encodes the relevant information into a mathematical model. Subsequently, an optimal control framework extracts the most efficient targets to steer the model into desired changes via external stimuli. The DataXflow software framework integrates three software pipelines, D2D for model fitting, a framework solving optimal control problems including external stimuli and JimenaE providing graphical user interfaces to employ the other frameworks lowering the barriers for the need of programming skills, and simultaneously automating reoccurring modeling tasks. Such tasks include equation generation from a graph and script generation allowing also to approach systems with many agents, like complex gene regulatory networks. A desired state of the model is defined, and therapeutic interventions are modeled as external stimuli. The optimal control framework purposefully exploits the model-encoded information by providing those external stimuli that effect the desired changes most efficiently. The implementation of DataXflow is available under https://github.com/MarvelousHopefull/DataXflow. We showcase its application by detecting specific drug targets for a therapy of lung cancer from measurement data to lower proliferation and increase apoptosis. By an iterative modeling process refining the topology of the model, the regulatory network of the tumor is generated from the data. An application of the optimal control framework in our example reveals the inhibition of AURKA and the activation of CDH1 as the most efficient drug target combination. DataXflow paves the way to an agile interplay between data generation and its analysis potentially accelerating cancer research by an efficient drug target identification, even in complex networks.

An orchestra of machine learning methods reveals landmarks in single-cell data exemplified with aging fibroblasts

In this work, a Python framework for characteristic feature extraction is developed and applied to gene expression data of human fibroblasts. Unlabeled feature selection objectively determines groups and minimal gene sets separating groups. ML explainability methods transform the features correlating with phenotypic differences into causal reasoning, supported by further pipeline and visualization tools, allowing user knowledge to boost causal reasoning. The purpose of the framework is to identify characteristic features that are causally related to phenotypic differences of single cells. The pipeline consists of several data science methods enriched with purposeful visualization of the intermediate results in order to check them systematically and infuse the domain knowledge about the investigated process. A specific focus is to extract a small but meaningful set of genes to facilitate causal reasoning for the phenotypic differences. One application could be drug target identification. For this purpose, the framework follows different steps: feature reduction (PFA), low dimensional embedding (UMAP), clustering ((H)DBSCAN), feature correlation (chi-square, mutual information), ML validation and explainability (SHAP, tree explainer). The pipeline is validated by identifying and correctly separating signature genes associated with aging in fibroblasts from single-cell gene expression measurements: PLK3, polo-like protein kinase 3; CCDC88A, Coiled-Coil Domain Containing 88A; STAT3, signal transducer and activator of transcription-3; ZNF7, Zinc Finger Protein 7; SLC24A2, solute carrier family 24 member 2 and lncRNA RP11-372K14.2. The code for the preprocessing step can be found in the GitHub repository https://github.com/AC-PHD/NoLabelPFA, along with the characteristic feature extraction https://github.com/LauritzR/characteristic-feature-extraction.