Molecular subtyping improves breast cancer diagnosis in the Copenhagen Breast Cancer Genomics Study.

BACKGROUNDIntrinsic molecular subtypes define distinct biological breast cancers and can be used to further improve diagnosis and risk allocation.METHODSThe Copenhagen Breast Cancer Genomics Study (CBCGS) prospectively included women diagnosed with breast cancer at Rigshospitalet from 2014 to 2021. Eligible patients were females with a primary invasive breast cancer (T1c, if N0M0; otherwise, any T, any N, or any M stage) and no prior malignancy. All patients underwent molecular profiling with the CIT256 and PAM50 molecular profile.RESULTSIn the study period, 2,816 patients were included in the CBCGS. Molecular subtyping showed an increase in nonluminal (molecular-apocrine, luminal C, and Basal-like) as compared with luminal (luminal A, luminal B, and Normal-like) subtypes with increasing stage from I to IV. Across all stages, we found a significant difference in survival among subtypes; 91% of patients with LumA were alive at 5 years compared with 91% for LumB, 84% for LumC, 82% for mApo, and 80% for Basal-like. We identified 442 tumors (16%) that were discordant in subtype between CIT256 and IHC. Discordant subtype proved to be a risk factor of death among patients with IHC luminal breast cancer (hazard ratio [HR], 2.08; 95% CI, 1.51-2.86) in a multivariable Cox regression analysis. Discordance occurred more often among patients with N3, stage IV, or grade III disease.CONCLUSIONOur findings indicate that molecular subtypes are a predominant classification for survival. Assessment is particularly crucial for patients with IHC luminal breast cancer with known high-risk factors, since they are at an increased risk of harboring an aggressive molecular subtype.

Whole genome sequencing in clinical practice.

Whole genome sequencing (WGS) is becoming the preferred method for molecular genetic diagnosis of rare and unknown diseases and for identification of actionable cancer drivers. Compared to other molecular genetic methods, WGS captures most genomic variation and eliminates the need for sequential genetic testing. Whereas, the laboratory requirements are similar to conventional molecular genetics, the amount of data is large and WGS requires a comprehensive computational and storage infrastructure in order to facilitate data processing within a clinically relevant timeframe. The output of a single WGS analyses is roughly 5 MIO variants and data interpretation involves specialized staff collaborating with the clinical specialists in order to provide standard of care reports. Although the field is continuously refining the standards for variant classification, there are still unresolved issues associated with the clinical application. The review provides an overview of WGS in clinical practice - describing the technology and current applications as well as challenges connected with data processing, interpretation and clinical reporting.

Neoepitope load, T cell signatures and PD-L2 as combined biomarker strategy for response to checkpoint inhibition immunotherapy.

Immune checkpoint inhibition for the treatment of cancer has provided a breakthrough in oncology, and several new checkpoint inhibition pathways are currently being investigated regarding their potential to provide additional clinical benefit. However, only a fraction of patients respond to such treatment modalities, and there is an urgent need to identify biomarkers to rationally select patients that will benefit from treatment. In this study, we explore different tumor associated characteristics for their association with favorable clinical outcome in a diverse cohort of cancer patients treated with checkpoint inhibitors. We studied 29 patients in a basket trial comprising 12 different tumor types, treated with 10 different checkpoint inhibition regimens. Our analysis revealed that even across this diverse cohort, patients achieving clinical benefit had significantly higher neoepitope load, higher expression of T cell signatures, and higher PD-L2 expression, which also correlated with improved progression-free and overall survival. Importantly, the combination of biomarkers serves as a better predictor than each of the biomarkers alone. Basket trials are frequently used in modern immunotherapy trial design, and here we identify a set of biomarkers of potential relevance across multiple cancer types, allowing for the selection of patients that most likely will benefit from immune checkpoint inhibition.

Ovarian cancer pathology characteristics as predictors of variant pathogenicity in BRCA1 and BRCA2.

BACKGROUND: The distribution of ovarian tumour characteristics differs between germline BRCA1 and BRCA2 pathogenic variant carriers and non-carriers. In this study, we assessed the utility of ovarian tumour characteristics as predictors of BRCA1 and BRCA2 variant pathogenicity, for application using the American College of Medical Genetics and the Association for Molecular Pathology (ACMG/AMP) variant classification system. METHODS: Data for 10,373 ovarian cancer cases, including carriers and non-carriers of BRCA1 or BRCA2 pathogenic variants, were collected from unpublished international cohorts and consortia and published studies. Likelihood ratios (LR) were calculated for the association of ovarian cancer histology and other characteristics, with BRCA1 and BRCA2 variant pathogenicity. Estimates were aligned to ACMG/AMP code strengths (supporting, moderate, strong). RESULTS: No histological subtype provided informative ACMG/AMP evidence in favour of BRCA1 and BRCA2 variant pathogenicity. Evidence against variant pathogenicity was estimated for the mucinous and clear cell histologies (supporting) and borderline cases (moderate). Refined associations are provided according to tumour grade, invasion and age at diagnosis. CONCLUSIONS: We provide detailed estimates for predicting BRCA1 and BRCA2 variant pathogenicity based on ovarian tumour characteristics. This evidence can be combined with other variant information under the ACMG/AMP classification system, to improve classification and carrier clinical management.

Extensive genomic analysis in patients with KRAS-mutated solid tumors shows high frequencies of concurrent alterations and potential targets but has limited clinical impact.

BACKGROUND: This study aimed to investigate the distribution and frequency of concurrent alterations in different cancers across KRAS subtypes and in different KRAS subtypes across cancers, and to identify potentially actionable targets and patients who received targeted treatment matched to their genomic profile (GP). MATERIALS AND METHODS: In this descriptive and single-center study, we included 188 patients with solid tumors harboring KRAS mutations in codon 12, 13, 61, 117, or 146, referred to the Phase 1 Unit, Rigshospitalet, Copenhagen, Denmark from mid-2016 to 2020. Genomic co-alterations were detected with whole-exome sequencing, RNA sequencing, SNP array, and mRNA expression array on fresh biopsies. The study is part of the Copenhagen Prospective Personalized Oncology study (NCT02290522). RESULTS: The majority of patients had colorectal cancer (60.1%), non-small cell lung cancer (11.2%), or pancreatic cancer (10.6%). Most tumors were KRAS-mutated in codon 12 or 13 (93.7%) including G12D (27.1%), G12V (26.6%), G12C (11.7%), and G13D (11.2%). A total of 175 different co-alterations were found, most frequently pathogenic APC and TP53 mutations (55.9% and 46.4%, respectively) and high expression of CEACAM5 (73.4%). Different cancers and KRAS subtypes showed different patterns of co-alterations, and 157 tumors (83.5%) had potentially actionable targets with varying evidence of targetability (assessed using ESMO Scale for Clinical Actionability of molecular Targets). Of the 188 patients included in the study, 15 (7.4%) received treatment matched to their GP (e.g., immunotherapy and synthetic lethality drugs), of whom one had objective partial response according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. CONCLUSION: Performing extensive genomic analysis in patients with known KRAS-mutated solid tumors may contribute with information to the genomic landscape of cancers and identify targets for immunotherapy or synthetic lethality drugs, but currently appears to have overall limited clinical impact, as few patients received targeted therapy matched to their GP.

Benchmarking full-length transcript single cell mRNA sequencing protocols.

BACKGROUND: Single cell mRNA sequencing technologies have transformed our understanding of cellular heterogeneity and identity. For sensitive discovery or clinical marker estimation where high transcript capture per cell is needed only plate-based techniques currently offer sufficient resolution. RESULTS: Here, we present a performance evaluation of four different plate-based scRNA-seq protocols. Our evaluation is aimed towards applications taxing high gene detection sensitivity, reproducibility between samples, and minimum hands-on time, as is required, for example, in clinical use. We included two commercial kits, NEBNext® Single Cell/ Low Input RNA Library Prep Kit (NEB®), SMART-seq® HT kit (Takara®), and the non-commercial protocols Genome & Transcriptome sequencing (G&T) and SMART-seq3 (SS3). G&T delivered the highest detection of genes per single cell. SS3 presented the highest gene detection per single cell at the lowest price. Takara® kit presented similar high gene detection per single cell, and high reproducibility between samples, but at the absolute highest price. NEB® delivered a lower detection of genes but remains an alternative to more expensive commercial kits. CONCLUSION: For the tested kits we found that ease-of-use came at higher prices. Takara can be selected for its ease-of-use to analyse a few samples, but we recommend the cheaper G&T-seq or SS3 for laboratories where a substantial sample flow can be expected.

MicroRNA-9-3p: a novel predictor of neurological outcome after cardiac arrest.

AIMS: Resuscitated out-of-hospital cardiac arrest (OHCA) patients who remain comatose after hospital arrival are at high risk of mortality due to anoxic brain injury. MicroRNA are small-non-coding RNA molecules ultimately involved in gene-silencing. They show promise as biomarkers, as they are stable in body fluids. The microRNA 9-3p (miR-9-3p) is associated with neurological injury in trauma and subarachnoid haemorrhage. METHODS AND RESULTS: This post hoc analysis considered all 171 comatose OHCA patients from a single centre in the target temperature management (TTM) trial. Patients were randomized to TTM at either 33°C or 36°C for 24 h. MicroRNA-9-3p (miR-9-3p) was measured in plasma sampled at admission and at 28, 48, and 72 h. There were no significant differences in age, gender, and pre-hospital data, including lactate level at admission, between miR-9-3p level quartiles. miR-9-3p levels changed markedly following OHCA with a peak at 48 h. Median miR-9-3p levels between TTM 33°C vs. 36°C were not different at any of the four time points. Elevated miR-9-3p levels at 48 h were strongly associated with an unfavourable neurological outcome [OR: 2.21, 95% confidence interval (CI): 1.64-3.15, P < 0.0001). MiR-9-3p was inferior to neuron-specific enolase in predicting functional neurological outcome [area under the curve: 0.79 (95% CI: 0.71-0.87) vs. 0.91 (95% CI: 0.85-0.97)]. CONCLUSION: MiR-9-3p is strongly associated with neurological outcome following OHCA, and the levels of miR-9-3p are peaking 48 hours following cardiac arrest.

Differences in gene expression despite identical histomorphology in sinonasal intestinal-type adenocarcinoma and metastases from colorectal adenocarcinoma.

Sinonasal intestinal-type adenocarcinoma (sITAC) is histomorphologically indistinguishable from colorectal adenocarcinoma (CRC) leading to diagnostic challenges. Metastases from CRCs to the sinonasal tract have been reported. The aim of the study was to identify a biomarker making it possible to distinguish between sITAC and metastases of colorectal origin. Formalin-fixated paraffin-embedded (FFPE) tissue from 20 consecutive patients with sITAC treated at Rigshospitalet, Denmark from 2005 to 2017, 20 patients with CRC, and second patients with both sinonasal and colorectal carcinomas were included, and RNA-sequencing was performed on all samples. Moreover, a series of 26 samples from metastasizing CRC were included (in-house data). 3139 differentially expressed genes were identified, of these several were deemed as possible biomarkers, including CSDE1, for which immunohistochemical staining was performed. sITAC and CRC differ in genomic expression. CSDE1, previously found upregulated in CRC, was significantly differentially expressed. Using immunohistochemical staining, no sITACs displayed strong and diffuse staining for CSDE1, which represents a potential marker to use in distinguishing sITAC from a metastasis of colorectal origin. This knowledge could improve the diagnostic process and hopefully the outcome in patients with this rare tumor.

Targeting the tumor mutanome for personalized vaccination in a TMB low non-small cell lung cancer.

BACKGROUND: Cancer is characterized by an accumulation of somatic mutations, of which a significant subset can generate cancer-specific neoepitopes that are recognized by autologous T cells. Such neoepitopes are emerging as important targets for cancer immunotherapy, including personalized cancer vaccination strategies. METHODS: We used whole-exome and RNA sequencing analysis to identify potential neoantigens for a patient with non-small cell lung cancer. Thereafter, we assessed the autologous T-cell reactivity to the candidate neoantigens using a long peptide approach in a cultured interferon gamma ELISpot and tracked the neoantigen-specific T-cells in the tumor by T-cell receptor (TCR) sequencing. In parallel, identified gene variants were incorporated into a Modified Vaccinia Ankara-based vaccine, which was evaluated in the human leucocyte antigen A*0201 transgenic mouse model (HHD). RESULTS: Sequencing revealed a tumor with a low mutational burden: 2219 sequence variants were identified from the primary tumor, of which 23 were expressed in the transcriptome, involving 18 gene products. We could demonstrate spontaneous T-cell responses to 5/18 (28%) mutated gene variants, and further analysis of the TCR repertoire of neoantigen-specific CD4+ and CD8+ T cells revealed TCR clonotypes that were expanded in both blood and tumor tissue. Following vaccination of HHD mice, de novo T-cell responses were generated to 4/18 (22%) mutated gene variants; T cells reactive against two variants were also evident in the autologous setting. Subsequently, we determined the major histocompatibility complex restriction of the T-cell responses and used in silico prediction tools to determine the likely neoepitopes. CONCLUSIONS: Our study demonstrates the feasibility of efficiently identifying tumor-specific neoantigens that can be targeted by vaccination in tumors with a low mutational burden, promising successful clinical exploitation, with trials currently underway.

A catalog of curated breast cancer genes.

PURPOSE: Decades of research have identified multiple genetic variants associated with breast cancer etiology. However, there is no database that archives breast cancer genes and variants responsible for predisposition. We set out to build a dynamic repository of curated breast cancer genes. METHODS: A comprehensive literature search was performed in PubMed and Google Scholar, followed by data extraction and harmonization for downstream analysis. RESULTS: Using a subset of 345 studies, we cataloged 652 breast cancer-associated loci across the genome. A majority of these were present in the non-coding region (i.e., intergenic (101) and intronic (345)), whereas only 158 were located within an exon. Using the odds ratio, we identified 429 loci to increase the disease risk and 198 to confer protection against breast cancer, whereas 25 were identified to both increase disease risk and confer protection against breast cancer. Chromosomal ideogram analysis indicated that chromosomes 17 and 19 have the highest density of breast cancer loci. We manually annotated and collated breast cancer genes in which a previous association between rare-monogenic variant and breast cancer has been documented. Finally, network and functional enrichment analysis revealed that steroid metabolism and DNA repair pathways were predominant among breast cancer genes and variants. CONCLUSIONS: We have built an online interactive catalog of curated breast cancer genes ( https://cbcg.dk ). This will expedite clinical diagnostics and support the ongoing efforts in managing breast cancer etiology. Moreover, the database will serve as an essential repository when designing new breast cancer multigene panels.

The Incidence, Survival, and HPV Impact of Second Primary Cancer following Primary Oropharyngeal Squamous Cell Carcinoma: A 20-Year Retrospective and Population-Based Study.

Second primary cancer (SPC) is the second most common cause of death among patients diagnosed with head and neck cancer. This study examined the risk of SPC following oropharyngeal squamous cell carcinoma (OPSCC) and the impact of human papillomavirus (HPV) on survival following SPC. The study was a population-based, retrospective study including all patients diagnosed with OPSCC in eastern Denmark from 2000-2020 who received curative intended treatment. The incidence rate ratio (IRR), age-adjusted incidence rates (AAIR), and hazard ratios (HR) were calculated. A total of 2584 patients with primary OPSCC were included (median follow-up time: 3.1 years), with 317 patients (12.3%) diagnosed with SPC. The risk of SPC was approximately five times the occurrence of cancer in the general population (IRR: 4.96). The median time to SPC after a primary OPSCC was 2.0 years (interquartile range (IQR) = 0.6-4.2 years). HPV-positive (HPV+) patients had a significantly longer median time to SPC, and a significant better survival compared to HPV-negative (HPV-) patients. SPC was most frequently found in lungs, head, and neck (LHN) for HPV- OPSCC patients and lungs followed by gender-specific (prostate, ovaries, or endometrium) for HPV+ OPSCC. There was a significant difference between the two groups when distributed between "within" or "outside" LHN. Patients with SPC outside LHN had a significant better overall survival. This knowledge should be considered during post-treatment surveillance and might guide targeted imaging.

A Comparison of Tools for Copy-Number Variation Detection in Germline Whole Exome and Whole Genome Sequencing Data.

Copy-number variations (CNVs) have important clinical implications for several diseases and cancers. Relevant CNVs are hard to detect because common structural variations define large parts of the human genome. CNV calling from short-read sequencing would allow single protocol full genomic profiling. We reviewed 50 popular CNV calling tools and included 11 tools for benchmarking in a reference cohort encompassing 39 whole genome sequencing (WGS) samples paired current clinical standard-SNP-array based CNV calling. Additionally, for nine samples we also performed whole exome sequencing (WES), to address the effect of sequencing protocol on CNV calling. Furthermore, we included Gold Standard reference sample NA12878, and tested 12 samples with CNVs confirmed by multiplex ligation-dependent probe amplification (MLPA). Tool performance varied greatly in the number of called CNVs and bias for CNV lengths. Some tools had near-perfect recall of CNVs from arrays for some samples, but poor precision. Several tools had better performance for NA12878, which could be a result of overfitting. We suggest combining the best tools also based on different methodologies: GATK gCNV, Lumpy, DELLY, and cn.MOPS. Reducing the total number of called variants could potentially be assisted by the use of background panels for filtering of frequently called variants.

Genomic Alterations in Human Papillomavirus-Positive and -Negative Conjunctival Squamous Cell Carcinomas.

Purpose: The genomic alterations contributing to the pathogenesis of conjunctival squamous cell carcinomas (SCCs) and their precursor lesions are poorly understood and hamper our ability to develop molecular therapies to reduce the recurrence rates and treatment-related morbidities of this disease. We aimed to characterize the somatic DNA alterations in human papillomavirus (HPV)-positive and HPV-negative conjunctival SCC. Methods: Patients diagnosed with conjunctival SCC in situ or SCC treated in ocular oncology referral centers in Denmark were included. HPV detection (HPV DNA PCR, p16 immunohistochemistry, and mRNA in situ hybridization) and targeted capture-based next-generation sequencing of 523 genes frequently involved in cancer were performed to describe the mutational profile based on HPV status. Results: Tumor tissue was available in 33 cases (n = 8 conjunctival SCCs in situ, n = 25 conjunctival SCCs), constituting 25 male and 8 female patients. Nine cases were HPV positive. The HPV-positive SCCs in situ and SCCs were characterized by transcriptionally active high-risk HPV (types 16 and 39) within the tumor cells, frequent mutations in PIK3CA (n = 5/9), and wild-type TP53, CDKN2A, and RB1, while the HPV-negative counterparts harbored frequent mutations in TP53 (n = 21/24), CDKN2A (n = 7/24), and RB1 (n = 6/24). Conclusions: Our findings have delineated two potentially distinct distributions of somatic mutations in conjunctival SCC based on HPV status-pointing to different biological mechanisms of carcinogenesis. The present findings support a causal role of HPV in a subset of conjunctival SCC.

Unveiling mRNP composition by fluorescence correlation and cross-correlation spectroscopy using cell lysates.

Understanding the mRNA life cycle requires information about the dynamics and macromolecular composition and stoichiometry of mRNPs. Fluorescence correlation and cross-correlation spectroscopy (FCS and FCCS) are appealing technologies to study these macromolecular structures because they have single molecule sensitivity and readily provide information about their molecular composition and dynamics. Here, we demonstrate how FCS can be exploited to study cytoplasmic mRNPs with high accuracy and reproducibility in cell lysates. Cellular lysates not only recapitulate data from live cells but provide improved readings and allow investigation of single mRNP analysis under particular conditions or following enzymatic treatments. Moreover, FCCS employing minute amounts of cells closely corroborated previously reported RNA dependent interactions and provided estimates of the relative overlap between factors in the mRNPs, thus depicting their heterogeneity. The described lysate-based FCS and FCCS analysis may not only complement current biochemical approaches but also provide novel opportunities for the quantitative analysis of the molecular composition and dynamics of single mRNPs.

A simple, safe and sensitive method for SARS-CoV-2 inactivation and RNA extraction for RT-qPCR.

The SARS-CoV-2 pandemic has created an urgent need for diagnostic tests to detect viral RNA. Commercial RNA extraction kits are often expensive, in limited supply, and do not always fully inactivate the virus. Together, this calls for the development of safer methods for SARS-CoV-2 extraction that utilize readily available reagents and equipment present in most standard laboratories. We optimized and simplified a RNA extraction method combining a high molar acidic guanidinium isothiocyanate (GITC) solution, phenol and chloroform. First, we determined the GITC/RNA dilution thresholds compatible with an efficient two-step RT-qPCR for B2M mRNA in nasopharyngeal (NP) or oropharyngeal (OP) swab samples. Second, we optimized a one-step RT-qPCR against SARS-CoV-2 using NP and OP samples. We furthermore tested a SARS-CoV-2 dilution series to determine the detection threshold. The method enables downstream detection of SARS-CoV-2 by RT-qPCR with high sensitivity (~4 viral RNA copies per RT-qPCR). The protocol is simple, safe, and expands analysis capacity as the inactivated samples can be used in RT-qPCR detection tests at laboratories not otherwise classified for viral work. The method takes about 30 min from swab to PCR-ready viral RNA and circumvents the need for commercial RNA purification kits.

Clinical implications of intrinsic molecular subtypes of breast cancer for sentinel node status.

Axillary lymph node status is an important prognostic factor for breast cancer patients and sentinel lymph node biopsy (SLNB) is a less invasive surgical proxy. We examined if consecutively derived molecular subtypes from primary breast cancers provide additional predictive value for SLNB status. 1556 patients with a breast cancer > 10 mm underwent primary surgical procedure including SLNB and tumor specimens were assigned with a transcriptomics-based molecular subtype. 1020 patients had a negative sentinel node (SN) and 536 a positive. A significant association between tumor size and SN status (p < 0.0001) was found across all samples, but no association between size and SN status (p = 0.14) was found for BasL tumors. A BasL subtype was a predictor of an SN-negative status (p = 0.001, OR 0.58, 95% CI 0.38;0.90) and among the BasL, postmenopausal status was a predictor for SN-negative status (p = 0.01). Overall survival was significantly lower (p = 0.02) in patients with BasL tumors and a positive SN. Interestingly, we identified a significant correlation between hormone receptor activity and SN status within the BasL subtype. Taken together, molecular subtypes and hormone receptor activity of breast cancers add predictive value for SLNB status.

Major driver mutations are shared between sinonasal intestinal-type adenocarcinoma and the morphologically identical colorectal adenocarcinoma.

PURPOSE: The purpose of our study was to compare genomic changes in sinonasal intestinal-type adenocarcinoma (sITAC) and colorectal adenocarcinoma (CRC), as they are histomorphologically indistinguishable. This can cause diagnostic difficulties as sinonasal tumours initially diagnosed as sITAC may represent metastasis from CRC, a frequent cancer. Previous studies have not uncovered the underlying mechanism behind the histomorphological resemblance. METHODS/PATIENTS: Tissue samples from all consecutive patients with sITAC at our facility (20 patients) were compared to samples from 20 patients with CRC as well as samples from 2 patients with both CRC and sinonasal tumours. DNA sequencing was performed using Illumina TruSight Oncology 500 panel consisting of 523 cancer-associated genes. Frequent mutations were inspected manually using the Integrative Genomics Viewer. RESULTS: Several well-known cancer-associated genes were mutated in the CRC group, but also in the sinonasal ITAC group. These genes included APC mutated in 65% of the CRC group and 37% of the sinonasal ITAC group, and TP53 mutated in 65% of CRC samples and 58% of ITAC samples. These shared mutations may explain the histomorphological similarities. Successful DNA sequencing was performed on the colorectal sample from one of the two patients with both CRC and sinonasal tumour. Comparing mutations in these samples from one patient we have shown that the sinonasal tumour in all probability was a CRC metastasis. CONCLUSION: We have identified several genetic similarities between sITAC and CRC. This discovery brings us closer to understanding mechanisms behind the development of sITAC-and hopefully in the future targeted therapy.

Interpretable Autoencoders Trained on Single Cell Sequencing Data Can Transfer Directly to Data from Unseen Tissues.

Autoencoders have been used to model single-cell mRNA-sequencing data with the purpose of denoising, visualization, data simulation, and dimensionality reduction. We, and others, have shown that autoencoders can be explainable models and interpreted in terms of biology. Here, we show that such autoencoders can generalize to the extent that they can transfer directly without additional training. In practice, we can extract biological modules, denoise, and classify data correctly from an autoencoder that was trained on a different dataset and with different cells (a foreign model). We deconvoluted the biological signal encoded in the bottleneck layer of scRNA-models using saliency maps and mapped salient features to biological pathways. Biological concepts could be associated with specific nodes and interpreted in relation to biological pathways. Even in this unsupervised framework, with no prior information about cell types or labels, the specific biological pathways deduced from the model were in line with findings in previous research. It was hypothesized that autoencoders could learn and represent meaningful biology; here, we show with a systematic experiment that this is true and even transcends the training data. This means that carefully trained autoencoders can be used to assist the interpretation of new unseen data.