A differential diagnosis between uterine leiomyoma and leiomyosarcoma using transcriptome analysis.

BACKGROUND: The objective of this study was to estimate the accuracy of transcriptome-based classifier in differential diagnosis of uterine leiomyoma and leiomyosarcoma. We manually selected 114 normal uterine tissue and 31 leiomyosarcoma samples from publicly available transcriptome data in UCSC Xena as training/validation sets. We developed pre-processing procedure and gene selection method to sensitively find genes of larger variance in leiomyosarcoma than normal uterine tissues. Through our method, 17 genes were selected to build transcriptome-based classifier. The prediction accuracies of deep feedforward neural network (DNN), support vector machine (SVM), random forest (RF), and gradient boosting (GB) models were examined. We interpret the biological functionality of selected genes via network-based analysis using GeneMANIA. To validate the performance of trained model, we additionally collected 35 clinical samples of leiomyosarcoma and leiomyoma as a test set (18 + 17 as 1st and 2nd test sets). RESULTS: We discovered genes expressed in a highly variable way in leiomyosarcoma while these genes are expressed in a conserved way in normal uterine samples. These genes were mainly associated with DNA replication. As gene selection and model training were made in leiomyosarcoma and uterine normal tissue, proving discriminant of ability between leiomyosarcoma and leiomyoma is necessary. Thus, further validation of trained model was conducted in newly collected clinical samples of leiomyosarcoma and leiomyoma. The DNN classifier performed sensitivity 0.88, 0.77 (8/9, 7/9) while the specificity 1.0 (8/8, 8/8) in two test data set supporting that the selected genes in conjunction with DNN classifier are well discriminating the difference between leiomyosarcoma and leiomyoma in clinical sample. CONCLUSION: The transcriptome-based classifier accurately distinguished uterine leiomyosarcoma from leiomyoma. Our method can be helpful in clinical practice through the biopsy of sample in advance of surgery. Identification of leiomyosarcoma let the doctor avoid of laparoscopic surgery, thus it minimizes un-wanted tumor spread.

Classification and Functional Analysis between Cancer and Normal Tissues Using Explainable Pathway Deep Learning through RNA-Sequencing Gene Expression.

Deep learning has proven advantageous in solving cancer diagnostic or classification problems. However, it cannot explain the rationale behind human decisions. Biological pathway databases provide well-studied relationships between genes and their pathways. As pathways comprise knowledge frameworks widely used by human researchers, representing gene-to-pathway relationships in deep learning structures may aid in their comprehension. Here, we propose a deep neural network (PathDeep), which implements gene-to-pathway relationships in its structure. We also provide an application framework measuring the contribution of pathways and genes in deep neural networks in a classification problem. We applied PathDeep to classify cancer and normal tissues based on the publicly available, large gene expression dataset. PathDeep showed higher accuracy than fully connected neural networks in distinguishing cancer from normal tissues (accuracy = 0.994) in 32 tissue samples. We identified 42 pathways related to 32 cancer tissues and 57 associated genes contributing highly to the biological functions of cancer. The most significant pathway was G-protein-coupled receptor signaling, and the most enriched function was the G1/S transition of the mitotic cell cycle, suggesting that these biological functions were the most common cancer characteristics in the 32 tissues.

A Deep Learning Model for Cell Growth Inhibition IC50 Prediction and Its Application for Gastric Cancer Patients.

Heterogeneity in intratumoral cancers leads to discrepancies in drug responsiveness, due to diverse genomics profiles. Thus, prediction of drug responsiveness is critical in precision medicine. So far, in drug responsiveness prediction, drugs' molecular "fingerprints", along with mutation statuses, have not been considered. Here, we constructed a 1-dimensional convolution neural network model, DeepIC50, to predict three drug responsiveness classes, based on 27,756 features including mutation statuses and various drug molecular fingerprints. As a result, DeepIC50 showed better cell viability IC50 prediction accuracy in pan-cancer cell lines over two independent cancer cell line datasets. Gastric cancer (GC) is not only one of the lethal cancer types in East Asia, but also a heterogeneous cancer type. Currently approved targeted therapies in GC are only trastuzumab and ramucirumab. Responsive GC patients for the drugs are limited, and more drugs should be developed in GC. Due to the importance of GC, we applied DeepIC50 to a real GC patient dataset. Drug responsiveness prediction in the patient dataset by DeepIC50, when compared to the other models, were comparable to responsiveness observed in GC cell lines. DeepIC50 could possibly accurately predict drug responsiveness, to new compounds, in diverse cancer cell lines, in the drug discovery process.

Personalized identification of altered pathways in cancer using accumulated normal tissue data.

MOTIVATION: Identifying altered pathways in an individual is important for understanding disease mechanisms and for the future application of custom therapeutic decisions. Existing pathway analysis techniques are mainly focused on discovering altered pathways between normal and cancer groups and are not suitable for identifying the pathway aberrance that may occur in an individual sample. A simple way to identify individual's pathway aberrance is to compare normal and tumor data from the same individual. However, the matched normal data from the same individual are often unavailable in clinical situation. Therefore, we suggest a new approach for the personalized identification of altered pathways, making special use of accumulated normal data in cases when a patient's matched normal data are unavailable. The philosophy behind our method is to quantify the aberrance of an individual sample's pathway by comparing it with accumulated normal samples. We propose and examine personalized extensions of pathway statistics, overrepresentation analysis and functional class scoring, to generate individualized pathway aberrance score. RESULTS: Collected microarray data of normal tissue of lung and colon mucosa are served as reference to investigate a number of cancer individuals of lung adenocarcinoma (LUAD) and colon cancer, respectively. Our method concurrently captures known facts of cancer survival pathways and identifies the pathway aberrances that represent cancer differentiation status and survival. It also provides more improved validation rate of survival-related pathways than when a single cancer sample is interpreted in the context of cancer-only cohort. In addition, our method is useful in classifying unknown samples into cancer or normal groups. Particularly, we identified 'amino acid synthesis and interconversion' pathway is a good indicator of LUAD (Area Under the Curve (AUC) 0.982 at independent validation). Clinical importance of the method is providing pathway interpretation of single cancer, even though its matched normal data are unavailable. AVAILABILITY AND IMPLEMENTATION: The method was implemented using the R software, available at our Web site: http://bibs.snu.ac.kr/ipas. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Practical approach to determine sample size for building logistic prediction models using high-throughput data.

An empirical method of sample size determination for building prediction models was proposed recently. Permutation method which is used in this procedure is a commonly used method to address the problem of overfitting during cross-validation while evaluating the performance of prediction models constructed from microarray data. But major drawback of such methods which include bootstrapping and full permutations is prohibitively high cost of computation required for calculating the sample size. In this paper, we propose that a single representative null distribution can be used instead of a full permutation by using both simulated and real data sets. During simulation, we have used a dataset with zero effect size and confirmed that the empirical type I error approaches to 0.05. Hence this method can be confidently applied to reduce overfitting problem during cross-validation. We have observed that pilot data set generated by random sampling from real data could be successfully used for sample size determination. We present our results using an experiment that was repeated for 300 times while producing results comparable to that of full permutation method. Since we eliminate full permutation, sample size estimation time is not a function of pilot data size. In our experiment we have observed that this process takes around 30min. With the increasing number of clinical studies, developing efficient sample size determination methods for building prediction models is critical. But empirical methods using bootstrap and permutation usually involve high computing costs. In this study, we propose a method that can reduce required computing time drastically by using representative null distribution of permutations. We use data from pilot experiments to apply this method for designing clinical studies efficiently for high throughput data.

Prediction of a time-to-event trait using genome wide SNP data.

BACKGROUND: A popular objective of many high-throughput genome projects is to discover various genomic markers associated with traits and develop statistical models to predict traits of future patients based on marker values. RESULTS: In this paper, we present a prediction method for time-to-event traits using genome-wide single-nucleotide polymorphisms (SNPs). We also propose a MaxTest associating between a time-to-event trait and a SNP accounting for its possible genetic models. The proposed MaxTest can help screen out nonprognostic SNPs and identify genetic models of prognostic SNPs. The performance of the proposed method is evaluated through simulations. CONCLUSIONS: In conjunction with the MaxTest, the proposed method provides more parsimonious prediction models but includes more prognostic SNPs than some naive prediction methods. The proposed method is demonstrated with real GWAS data.

Minoxidil Regulates Aging-Like Phenotypes in Rat Cortical Astrocytes In Vitro.

Mainly due to the slanted focus on the mechanism and regulation of neuronal aging, research on astrocyte aging and its modulation during brain aging is scarce. In this study, we established aged astrocyte culture model by long-term culturing. Cellular senescence was confirmed through SA-ß-gal staining as well as through the examination of morphological, molecular, and functional markers. RNA sequencing and functional analysis of astrocytes were performed to further investigate the detailed characteristics of the aged astrocyte model. Along with aged phenotypes, decreased astrocytic proliferation, migration, mitochondrial energetic function and support for neuronal survival and differentiation has been observed in aged astrocytes. In addition, increased expression of cytokines and chemokine-related factors including plasminogen activator inhibitor -1 (PAI-1) was observed in aged astrocytes. Using the RNA sequencing results, we searched potential drugs that can normalize the dysregulated gene expression pattern observed in long-term cultured aged astrocytes. Among several candidates, minoxidil, a pyrimidine-derived anti-hypertensive and anti-pattern hair loss drug, normalized the increased number of SA-ß-gal positive cells and nuclear size in aged astrocytes. In addition, minoxidil restored up-regulated activity of PAI-1 and increased mitochondrial superoxide production in aged astrocytes. We concluded that long term culture of astrocytes can be used as a reliable model for the study of astrocyte senescence and minoxidil can be a plausible candidate for the regulation of brain aging.

Enhanced Glutaminolysis Drives Hypoxia-Induced Chemoresistance in Pancreatic Cancer.

Pancreatic ductal adenocarcinoma (PDAC) exhibits severe hypoxia, which is associated with chemoresistance and worse patient outcome. It has been reported that hypoxia induces metabolic reprogramming in cancer cells. However, it is not well known whether metabolic reprogramming contributes to hypoxia. Here, we established that increased glutamine catabolism is a fundamental mechanism inducing hypoxia, and thus chemoresistance, in PDAC cells. An extracellular matrix component-based in vitro three-dimensional cell printing model with patient-derived PDAC cells that recapitulate the hypoxic status in PDAC tumors showed that chemoresistant PDAC cells exhibit markedly enhanced glutamine catabolism compared with chemoresponsive PDAC cells. The augmented glutamine metabolic flux increased the oxygen consumption rate via mitochondrial oxidative phosphorylation (OXPHOS), promoting hypoxia and hypoxia-induced chemoresistance. Targeting glutaminolysis relieved hypoxia and improved chemotherapy efficacy in vitro and in vivo. This work suggests that targeting the glutaminolysis-OXPHOS-hypoxia axis is a novel therapeutic target for treating patients with chemoresistant PDAC. SIGNIFICANCE: Increased glutaminolysis induces hypoxia via oxidative phosphorylation-mediated oxygen consumption and drives chemoresistance in pancreatic cancer, revealing a potential therapeutic strategy of combining glutaminolysis inhibition and chemotherapy to overcome resistance.

A transcriptome-Based Deep Neural Network Classifier for Identifying the Site of Origin in Mucinous Cancer.

Purpose: There is a lack of tools for identifying the site of origin in mucinous cancer. This study aimed to evaluate the performance of a transcriptome-based classifier for identifying the site of origin in mucinous cancer. Materials And Methods: Transcriptomic data of 1878 non-mucinous and 82 mucinous cancer specimens, with 7 sites of origin, namely, the uterine cervix (CESC), colon (COAD), pancreas (PAAD), stomach (STAD), uterine endometrium (UCEC), uterine carcinosarcoma (UCS), and ovary (OV), obtained from The Cancer Genome Atlas, were used as the training and validation sets, respectively. Transcriptomic data of 14 mucinous cancer specimens from a tissue archive were used as the test set. For identifying the site of origin, a set of 100 differentially expressed genes for each site of origin was selected. After removing multiple iterations of the same gene, 427 genes were chosen, and their RNA expression profiles, at each site of origin, were used to train the deep neural network classifier. The performance of the classifier was estimated using the training, validation, and test sets. Results: The accuracy of the model in the training set was 0.998, while that in the validation set was 0.939 (77/82). In the test set which is newly sequenced from a tissue archive, the model showed an accuracy of 0.857 (12/14). t-SNE analysis revealed that samples in the test set were part of the clusters obtained for the training set. Conclusion: Although limited by small sample size, we showed that a transcriptome-based classifier could correctly identify the site of origin of mucinous cancer.

Pirenperone relieves the symptoms of fragile X syndrome in Fmr1 knockout mice.

Fragile X syndrome (FXS) is a neurodevelopmental disorder that is caused by the loss of Fragile X-linked mental retardation protein (FMRP), an RNA binding protein that can bind and recognize different RNA structures and regulate the target mRNAs' translation involved in neuronal synaptic plasticity. Perturbations of this gene expression network have been related to abnormal behavioral symptoms such as hyperactivity, and impulsivity. Considering the roles of FMRP in the modulation of mRNA translation, we investigated the differentially expressed genes which might be targeted to revert to normal and ameliorate behavioral symptoms. Gene expression data was analyzed and used the connectivity map (CMap) to understand the changes in gene expression in FXS and predict the effective drug candidates. We analyzed the GSE7329 dataset that had 15 control and 8 FXS patients' lymphoblastoid samples. Among 924 genes, 42 genes were selected as signatures for CMap analysis, and 24 associated drugs were found. Pirenperone was selected as a potential drug candidate for FXS for its possible antipsychotic effect. Treatment of pirenperone increased the expression level of Fmr1 gene. Moreover, pirenperone rescued the behavioral deficits in Fmr1 KO mice including hyperactivity, spatial memory, and impulsivity. These results suggest that pirenperone is a new drug candidate for FXS, which should be verified in future studies.

Integrated analysis of ascites and plasma extracellular vesicles identifies a miRNA-based diagnostic signature in ovarian cancer.

Ovarian cancer is mostly diagnosed at advantaged stages due to the lack of early diagnostic biomarkers. The common metastasis pattern is characterized by peritoneal dissemination with a formation of malignant ascites. Extracellular vesicles (EVs) are emerging as promising clinical biomarkers in liquid biopsy. Here, we aimed to investigate robust liquid biopsy-based EV miRNA biomarkers for ovarian cancer diagnosis and metastasis regulation. EVs were isolated from malignant ascites and plasma of ovarian cancer patients as well as the benign control counterparts of patients with benign gynecologic diseases. EV small RNA sequencing identified a panel of eight miRNAs (miR-1246, miR-1290, miR-483, miR-429, miR-34b-3p, miR-34c-5p, miR-145-5p, miR-449a) based on dysregulated miRNAs overlapped in the ascites and plasma subset. The ovarian cancer EV miRNA (OCEM) signature developed based on these eight miRNAs demonstrated high diagnostic accuracy in our in-house dataset and multiple public datasets across diverse clinical samples (blood, tissue and urine). In addition, malignant ascites-derived EVs could significantly facilitate the aggressive property of ovarian cancer cells and boost the growth of ascites-derived organoids. Notably, miR-1246 and miR-1290 shuttled in malignant ascites-derived EVs were identified to promote the invasion and migration of ovarian cancer cells through regulating a common target RORα.

A fast algorithm for exact sequence search in biological sequences using polyphase decomposition.

MOTIVATION: Exact sequence search allows a user to search for a specific DNA subsequence in a larger DNA sequence or database. It serves as a vital block in many areas such as Pharmacogenetics, Phylogenetics and Personal Genomics. As sequencing of genomic data becomes increasingly affordable, the amount of sequence data that must be processed will also increase exponentially. In this context, fast sequence search algorithms will play an important role in exploiting the information contained in the newly sequenced data. Many existing algorithms do not scale up well for large sequences or databases because of their high-computational costs. This article describes an efficient algorithm for performing fast searches on large DNA sequences. It makes use of hash tables of Q-grams that are constructed after downsampling the database, to enable efficient search and memory use. Time complexity for pattern search is reduced using beam pruning techniques. Theoretical complexity calculations and performance figures are presented to indicate the potential of the proposed algorithm.

Deep Learning for Integrated Analysis of Insulin Resistance with Multi-Omics Data.

Technological advances in next-generation sequencing (NGS) have made it possible to uncover extensive and dynamic alterations in diverse molecular components and biological pathways across healthy and diseased conditions. Large amounts of multi-omics data originating from emerging NGS experiments require feature engineering, which is a crucial step in the process of predictive modeling. The underlying relationship among multi-omics features in terms of insulin resistance is not well understood. In this study, using the multi-omics data of type II diabetes from the Integrative Human Microbiome Project, from 10,783 features, we conducted a data analytic approach to elucidate the relationship between insulin resistance and multi-omics features, including microbiome data. To better explain the impact of microbiome features on insulin classification, we used a developed deep neural network interpretation algorithm for each microbiome feature's contribution to the discriminative model output in the samples.

Effects of Intraperitoneal N-methyl-D-aspartate (NMDA) Administration on Nociceptive/Repetitive Behaviors in Juvenile Mice.

Dysregulation of excitatory neurotransmission has been implicated in the pathogenesis of neuropsychiatric disorders. Pharmacological inhibition of N-methyl-D-aspartate (NMDA) receptors is widely used to model neurobehavioral pathologies and underlying mechanisms. There is ample evidence that overstimulation of NMDA-dependent neurotransmission may induce neurobehavioral abnormalities, such as repetitive behaviors and hypersensitization to nociception and cognitive disruption, pharmacological modeling using NMDA has been limited due to the induction of neurotoxicity and blood brain barrier breakdown, especially in young animals. In this study, we examined the effects of intraperitoneal NMDA-administration on nociceptive and repetitive behaviors in ICR mice. Intraperitoneal injection of NMDA induced repetitive grooming and tail biting/licking behaviors in a dose- and age-dependent manner. Nociceptive and repetitive behaviors were more prominent in juvenile mice than adult mice. We did not observe extensive blood brain barrier breakdown or neuronal cell death after peritoneal injection of NMDA, indicating limited neurotoxic effects despite a significant increase in NMDA concentration in the cerebrospinal fluid. These findings suggest that the observed behavioral changes were not mediated by general NMDA toxicity. In the hot plate test, we found that the latency of paw licking and jumping decreased in the NMDA-exposed mice especially in the 75 mg/kg group, suggesting increased nociceptive sensitivity in NMDA-treated animals. Repetitive behaviors and increased pain sensitivity are often comorbid in psychiatric disorders (e.g., autism spectrum disorder). Therefore, the behavioral characteristics of intraperitoneal NMDA-administered mice described herein may be valuable for studying the mechanisms underlying relevant disorders and screening candidate therapeutic molecules.

Storing Digital Information in the Long Read DNA.

There is urgent need in effective and cost-efficient data storage as worldwide requirement of data storage rapidly growing. DNA has introduced a new tool for storing digital information. Recent studies successfully store digital information such as text and gif animation. Previous studies tackle technical hurdles due to errors from DNA synthesis and sequencing. Studies also have focused on the strategy which makes use of 100-150 bps of read size in both synthesis and sequencing. In this paper, we suggest novel data encoding / decoding scheme which makes use of long read DNA (~1,000bp). This enables accurate recovery of stored digital information with a smaller number of reads than previous approach. Also, the approach reduces sequencing time.

Ovarian Cancer Prognostic Prediction Model Using RNA Sequencing Data.

Ovarian cancer is one of the leading causes of cancer-related deaths in gynecologic malignancies. Over 70 % of ovarian cancer cases are high-grade serous ovarian cancers (HGSC) and have high death rates due to their resistance to chemotherapy. Despite advances in surgical and pharmaceutical therapies, overall survival rates are not good and accurate prediction of prognosis is not easy because of the highly heterogeneous nature of ovarian cancer. To improve patient's prognosis through proper treatment, we present a prognostic prediction model by integrating the high dimensional RNA sequencing data with their clinical data through the following steps: (1) gene filtration, (2) pre-screening, (3) gene marker selection (4) integrated study of selected gene markers and prediction model building. These steps of the prognostic prediction model can be applied to other types of cancer besides ovarian cancer.

Fine-scale map of encyclopedia of DNA elements regions in the Korean population.

The International HapMap Project aims to generate detailed human genome variation maps by densely genotyping single-nucleotide polymorphisms (SNPs) in CEPH, Chinese, Japanese, and Yoruba samples. This will undoubtedly become an important facility for genetic studies of diseases and complex traits in the four populations. To address how the genetic information contained in such variation maps is transferable to other populations, the Korean government, industries, and academics have launched the Korean HapMap project to genotype high-density Encyclopedia of DNA Elements (ENCODE) regions in 90 Korean individuals. Here we show that the LD pattern, block structure, haplotype diversity, and recombination rate are highly concordant between Korean and the two HapMap Asian samples, particularly Japanese. The availability of information from both Chinese and Japanese samples helps to predict more accurately the possible performance of HapMap markers in Korean disease-gene studies. Tagging SNPs selected from the two HapMap Asian maps, especially the Japanese map, were shown to be very effective for Korean samples. These results demonstrate that the HapMap variation maps are robust in related populations and will serve as an important resource for the studies of the Korean population in particular.

Acquired resistance to LY2874455 in FGFR2-amplified gastric cancer through an emergence of novel FGFR2-ACSL5 fusion.

BACKGROUND: Fibroblast growth factor 2 (FGFR2) amplification, occurring in ~2-9% of gastric cancers (GC), is associated with poor overall survival. RESULTS: RNA sequencing identified a novel FGFR2-ACSL5 fusion in the resistant tumor that was absent from the matched pre-treatment tumor. The FGFR2-amplified PDC line was sensitive to FGFR inhibitors whereas the PDC line with concomitant FGFR2 amplification and FGFR2-ACSL5 fusion exhibited resistance. Additionally, the FGFR2-amplified GC PDC line, which was initially sensitive to FGFR2 inhibitors, subsequently also developed resistance. MATERIALS AND METHODS: We identified an FGFR2-amplified patient with GC, who demonstrated a dramatic and long-term response to LY2874455, a pan-FGFR inhibitor, but eventually developed an acquired LY2874455 resistance. Following resistance development, an endoscopic biopsy was performed for transcriptome sequencing and patient-derived tumor cell line (PDC) establishment to elucidate the underlying molecular alterations. CONCLUSIONS: FGFR inhibitors may function against FGFR2-amplified GC, and a novel FGFR2-ACSL5 fusion identified by transcriptomic characterization may underlie clinically acquired resistance. IMPLICATIONS FOR PRACTICE: Poor treatment response represents a substantial concern in patients with gastric cancer carrying multiple FGFR2 gene copies. Here, we show the utility of a general FGFR inhibitor for initial response prior to treatment resistance and report the first characterization of a potential resistance mechanism involving an FGFR2-ACSL5 fusion protein.

LY-6K gene: a novel molecular marker for human breast cancer.

A full-length cDNA was identified using one STS sequence containing an SNP (single nucleotide polymorphism) derived from genomic DNAs of breast cancer patients using a variety of bioinformatics tools. The cDNA encodes LY-6K, a novel member protein of the Ly-6/uPAR superfamily. It has been annotated as a target antigen for the HNSCC (head-and neck squamous cell carcinoma). We isolated the LY-6K gene from genomic DNAs obtained from breast cancer patients through large scale, case-control-screening. We performed northern blot hybridization and semi-quantitative RT-PCR on a human multiple-tissue mRNA blot from several breast cancer patients. We investigated the expression level of the LY-6K gene in human breast cancer, and compared this to expression in human normal breast tissue. We found that LY-6K was more highly expressed in the mRNA of breast tumors compared to its expression in normal breast tissue. These results suggest that LY-6K is not only a target antigen for HNSCC but also a significant new molecular marker for diagnosis and gene therapy in patients with breast cancer.

Gene Expression Profiling of Breast Cancer Brain Metastasis.

The biology of breast cancer brain metastasis (BCBM) is poorly understood. We aimed to explore genes that are implicated in the process of brain metastasis of primary breast cancer (BC). NanoString nCounter Analysis covering 252 target genes was used for comparison of gene expression levels between 20 primary BCs that relapsed to brain and 41 BCBM samples. PAM50-based intrinsic subtypes such as HER2-enriched and basal-like were clearly over-represented in BCBM. A panel of 22 genes was found to be significantly differentially expressed between primary BC and BCBM. Five of these genes, CXCL12, MMP2, MMP11, VCAM1, and MME, which have previously been associated with tumor progression, angiogenesis, and metastasis, clearly discriminated between primary BC and BCBM. Notably, the five genes were significantly upregulated in primary BC compared to BCBM. Conversely, SOX2 and OLIG2 genes were upregulated in BCBM. These genes may participate in metastatic colonization but not in primary tumor development. Among patient-matched paired samples (n = 17), a PAM50 molecular subtype conversion was observed in eight cases (47.1%), with a trend toward unfavorable subtypes in patients with the distinct gene expression. Our findings, although not conclusive, reveal differentially expressed genes that might mediate the brain metastasis process.