Transcriptomic Harmonization as the Way for Suppressing Cross-Platform Bias and Batch Effect

A simplified machine learning model utilizing platelet-related genes for predicting poor prognosis in sepsis

Background

Thrombocytopenia is a known prognostic factor in sepsis, yet the relationship between platelet-related genes and sepsis outcomes remains elusive. We developed a machine learning (ML) model based on platelet-related genes to predict poor prognosis in sepsis. The model underwent rigorous evaluation on six diverse platforms, ensuring reliable and versatile findings.

Methods

A retrospective analysis of platelet data from 365 sepsis patients confirmed the predictive role of platelet count in prognosis. We employed COX analysis, Least Absolute Shrinkage and Selection Operator (LASSO) and Support Vector Machine (SVM) techniques to identify platelet-related genes from the GSE65682 dataset. Subsequently, these genes were trained and validated on six distinct platforms comprising 719 patients, and compared against the Acute Physiology and Chronic Health Evaluation II (APACHE II) and Sequential Organ-Failure Assessment (SOFA) score.

Results

A PLT count <100×109/L independently increased the risk of death in sepsis patients (OR = 2.523; 95% CI: 1.084-5.872). The ML model, based on five platelet-related genes, demonstrated impressive area under the curve (AUC) values ranging from 0.5 to 0.795 across various validation platforms. On the GPL6947 platform, our ML model outperformed the APACHE II score with an AUC of 0.795 compared to 0.761. Additionally, by incorporating age, the model's performance was further improved to an AUC of 0.812. On the GPL4133 platform, the initial AUC of the machine learning model based on five platelet-related genes was 0.5. However, after including age, the AUC increased to 0.583. In comparison, the AUC of the APACHE II score was 0.604, and the AUC of the SOFA score was 0.542.

Conclusion

Our findings highlight the broad applicability of this ML model, based on platelet-related genes, in facilitating early treatment decisions for sepsis patients with poor outcomes. Our study paves the way for advancements in personalized medicine and improved patient care.

Uniformly shaped harmonization combines human transcriptomic data from different platforms while retaining their biological properties and differential gene expression patterns

Introduction: Co-normalization of RNA profiles obtained using different experimental platforms and protocols opens avenue for comprehensive comparison of relevant features like differentially expressed genes associated with disease. Currently, most of bioinformatic tools enable normalization in a flexible format that depends on the individual datasets under analysis. Thus, the output data of such normalizations will be poorly compatible with each other. Recently we proposed a new approach to gene expression data normalization termed Shambhala which returns harmonized data in a uniform shape, where every expression profile is transformed into a pre-defined universal format. We previously showed that following shambhalization of human RNA profiles, overall tissue-specific clustering features are strongly retained while platform-specific clustering is dramatically reduced. Methods: Here, we tested Shambhala performance in retention of fold-change gene expression features and other functional characteristics of gene clusters such as pathway activation levels and predicted cancer drug activity scores. Results: Using 6,793 cancer and 11,135 normal tissue gene expression profiles from the literature and experimental datasets, we applied twelve performance criteria for different versions of Shambhala and other methods of transcriptomic harmonization with flexible output data format. Such criteria dealt with the biological type classifiers, hierarchical clustering, correlation/regression properties, stability of drug efficiency scores, and data quality for using machine learning classifiers. Discussion: Shambhala-2 harmonizer demonstrated the best results with the close to 1 correlation and linear regression coefficients for the comparison of training vs validation datasets and more than two times lesser instability for calculation of drug efficiency scores compared to other methods.

Large-scale assessment of pros and cons of autopsy-derived or tumor-matched tissues as the norms for gene expression analysis in cancers

Normal tissues are essential for studying disease-specific differential gene expression. However, healthy human controls are typically available only in postmortal/autopsy settings. In cancer research, fragments of pathologically normal tissue adjacent to tumor site are frequently used as the controls. However, it is largely underexplored how cancers can systematically influence gene expression of the neighboring tissues. Here we performed a comprehensive pan-cancer comparison of molecular profiles of solid tumor-adjacent and autopsy-derived "healthy" normal tissues. We found a number of systemic molecular differences related to activation of the immune cells, intracellular transport and autophagy, cellular respiration, telomerase activation, p38 signaling, cytoskeleton remodeling, and reorganization of the extracellular matrix. The tumor-adjacent tissues were deficient in apoptotic signaling and negative regulation of cell growth including G2/M cell cycle transition checkpoint. We also detected an extensive rearrangement of the chemical perception network. Molecular targets of 32 and 37 cancer drugs were over- or underexpressed, respectively, in the tumor-adjacent norms. These processes may be driven by molecular events that are correlated between the paired cancer and adjacent normal tissues, that mostly relate to inflammation and regulation of intracellular molecular pathways such as the p38, MAPK, Notch, and IGF1 signaling. However, using a model of macaque postmortal tissues we showed that for the 30 min - 24-hour time frame at 4ºC, an RNA degradation pattern in lung biosamples resulted in an artifact "differential" expression profile for 1140 genes, although no differences could be detected in liver. Thus, such concerns should be addressed in practice.