expHRD: an individualized, transcriptome-based prediction model for homologous recombination deficiency assessment in cancer

Abstract

Background: Homologous recombination deficiency (HRD) stands as a clinical indicator for discerning responsive outcomes to platinum-based chemotherapy and poly ADP-ribose polymerase (PARP) inhibitors. One of the conventional approaches to HRD prognostication has generally centered on identifying deleterious mutations within the BRCA1/2 genes, along with quantifying the genomic scars, such as Genomic Instability Score (GIS) estimation with scarHRD. However, the scarHRD method has limitations in scenarios involving tumors bereft of corresponding germline data. Although several RNA-seq-based HRD prediction algorithms have been developed, they mainly support cohort-wise classification, thereby yielding HRD status without furnishing an analogous quantitative metric akin to scarHRD. This study introduces the expHRD method, which operates as a novel transcriptome-based framework tailored to n-of-1-style HRD scoring.

Results: The prediction model has been established using the elastic net regression method in the Cancer Genome Atlas (TCGA) pan-cancer training set. The bootstrap technique derived the HRD geneset for applying the expHRD calculation. The expHRD demonstrated a notable correlation with scarHRD and superior performance in predicting HRD-high samples. We also performed intra- and extra-cohort evaluations for clinical feasibility in the TCGA-OV and the Genomic Data Commons (GDC) ovarian cancer cohort, respectively. The innovative web service designed for ease of use is poised to extend the realms of HRD prediction across diverse malignancies, with ovarian cancer standing as an emblematic example.

Conclusions: Our novel approach leverages the transcriptome data, enabling the prediction of HRD status with remarkable precision. This innovative method addresses the challenges associated with limited available data, opening new avenues for utilizing transcriptomics to inform clinical decisions.

Keywords: Bootstrap; Elastic net regression; Homologous recombination deficiency; Transcriptome; scarHRD.

Fig. 1

Schematic representation of transcriptome-based HRD prediction model development and validation. Schematic illustration describing the overall process of serial machine-learning training and validation for the development of an HRD prediction algorithm in the cancer transcriptome

Fig. 2

Evaluation of the HRD prediction model in the TCGA-pan cancer. a Cross-validation analysis of the machine-learning model. The x-axis denotes the number of elastic net cross-validation iterations. The left-y-axis signifies the count of features (genes), while the right-y axis indicates Pearson’s correlation coefficient (PCC) with the scarHRD score post-machine learning. Black closed circles linked by solid lines and white circles connected by dotted lines correspond to the gene count and PCC, respectively, across each cross-validation step. b Correlation pattern across TCGA-pan cancer cohorts. Bar graph depicting the PCC between the predicted HRD score and scarHRD score in the TCGA-pan cancer test set, encompassing various cancer types including KIRP (kidney renal clear papillary cell carcinoma), UCEC (uterine corpus endometrial carcinoma), BRCA (breast invasive carcinoma), KICH (kidney chromophobe), BLCA (bladder urothelial carcinoma), CESC (cervical squamous cell carcinoma and endocervical adenocarcinoma), OV (ovarian serous cystadenocarcinoma), STAD (stomach adenocarcinoma), SARC (sarcoma), UCS (uterine carcinosarcoma), LIHC (liver hepatocellular carcinoma), PRAD (prostate adenocarcinoma), LGG (brain lower grade glioma), TNBC (triplet negative breast cancer), HNSC (head and neck squamous cell carcinoma), MESO (mesothelioma), READ (rectum adenocarcinoma), SKCM (skin cutaneous melanoma), LUAD (lung adenocarcinoma), PAAD (pancreatic adenocarcinoma), ESCA (esophageal carcinoma), COAD (colon adenocarcinoma), KIRC (kidney renal clear cell carcinoma), ACC (adrenocortical carcinoma), LUSC (lung squamous cell carcinoma), THYM (thymoma), CHOL (cholangiocarcinoma), PCPG (pheochromocytoma and paraganglioma), GBM (glioblastoma multiforme), THCA (thyroid carcinoma), UVM (uveal melanoma), DLBC (lymphoid neoplasm diffuse large B-cell lymphoma), and TGCT (testicular germ cell tumours). Significance levels denoted as *, **, and *** indicate P-values < 0.05, < 0.001, and < 0.0001, respectively. The frequency of HRD (scarHRD score ≥ 42) in each tumour type is displayed. c Correlation between scarHRD and predicted HRD score (pHRD) in the TCGA-pan cancer test set. Pearson’s correlation-regression line was calculated, with the dark dotted line illustrating pan-cancer correlation and the red line representing TCGA-OV set correlation. The numeric number in each bar plot represents the frequency of HRD positive samples in cancer types. Frequency: the number of HRD positive sample / the number of sample with scarHRD score

Fig. 3

Validation of expHRD performance in the TCGA-OV test set. a Pearson’s correlation between expHRD and scarHRD in the TCGA-OV test set (n = 58, PCC = 0.768, p = 2.045e−12). The blue line denotes the regression line, while the shaded area represents the 95% confidence interval (CI). b Receiver operating characteristic (ROC) curve plotting sensitivity against 1-specificity values for expHRD score’s capacity to predict scarHRD-high instances within TCGA-OV test samples. c, d Kaplan–Meier overall survival analysis contrasting patients with high vs. low scarHRD (c) or expHRD (d) status within the TCGA-OV test set. P-values obtained via the log-rank test

Fig. 4

The web interface for expHRD calculation. The front page of the web interface for expHRD calculation demonstrates the query for uploading (a) and the input file overview (b) of the user’s gene-expression data

Fig. 5

The result page of the web interface. a The result page of the web interface exhibited upon selecting “Run” after file upload presents the HRD score graph with the TCGA-OV test sample’s regression function (a, left). The histogram illustrates the distribution of HRD scores (a, right). b The result table showcases the calculated expHRD alongside the predicted HRD (p_hrd), both derived from the expHRD outcomes. The values for lower_hrd and higher_hrd are determined by the 95% confidence intervals (CI) of the predicted HRD