Noncoding regions occupy more than 98% of the human genome, and almost all of them (> 93%) are actively transcribed [1–3]. These transcripts are dominated by long noncoding RNAs (lncRNAs), including lncRNAs, antisense RNAs, and pseudogenes in the present review. The transcription of lncRNAs requires a large amount of energy in the human genome. Theoretically, energy consumption generates a function to fit the low-energy law in a stable physical system. Therefore, all energy-consuming lncRNAs hypothetically carry a certain degree of biological function under given conditions, although most of these functions remain unknown. Understanding the functions of these lncRNAs is critical to understanding their functions in the big picture of the human genome.

Recent evidence suggests that lncRNAs can function as functional molecules, but a real breakthrough came with the recognition that lncRNAs play an influential role in processes such as growth and metabolism [4–7]. Since the breakthrough discovery of lncRNA functions, studies on lncRNA functions have exponentially increased [4–13], attempting to address basic concepts of their biology, such as transcription initiation and regulatory systems [14]. These lncRNA functional studies have conventionally adopted the conceptual framework of protein-based functional systems, which has provided exciting data to provide a preliminary picture of lncRNA function [14]. For example, protein-based polymerase II (Pol II) has been recognized as a primary enzyme that initiates lncRNA transcription [14]. Identification of lncRNAs has also been based on the concept of protein identification, using promoters, start codons, poly(A) tails, Pol II, and DNA conservation [15]. The GENCODE project V35 combined both messenger RNA (mRNA) concepts and sequencing approaches to identify 40,702 lncRNAs, which merged long intergenic noncoding RNAs (lincRNAs) and antisense RNAs [16]. The Functional ANnoTation Of the Mammalian genome (FANTOM) project employed the 5’ strategy to capture 5’ mRNA caps and identify 19,175 lncRNAs [17]. However, these lncRNA functional studies were conducted using conventional approaches that rely on individual experiments and limited datasets. This leads to unavoidably biased observations specific to biological conditions such as tissue types and genetic and epigenetic backgrounds. For example, 78% of lncRNAs collected by the GENCODE project V35 [16] and FANTOM project [17] are condition-dependent [1]. These condition-specific studies are unlikely to generate endogenous lncRNA patterns for understanding the general principles of lncRNA biology.

The concept of condition-dependent lncRNA function has condition-specific implications. It is not surprising that the roles of lncRNAs in cancer are tissue-specific, and lncRNAs have been recognized as secondary factors in tumorigenesis. Proteins are considered the most critical factors that regulate cancer progression [18–26]. Intense studies on proteins have generated a wealth of useful data for clinical treatment to extend the life span of cancer patients; however, these conventional studies have failed to uncover the endogenous mechanism of a common regulatory regime shared across all cancer types. The mechanisms underlying specific cancer types and subtypes have been emphasized in cancer research and therapy.

Fortunately, recent big data studies have significantly advanced our understanding of lncRNA functions and created a conceptual breakthrough in the lncRNA functional system [1, 27]. Analysis of multiple large datasets has demonstrated that lncRNAs are evolutionary drivers of animal lifespan across the animal kingdom by lowering energy consumption [28]. This explains why the human genome requires 98% of noncoding regions that perform broader functions than previously thought with low energy. Big data studies have further developed novel computational algorithms that can find endogenous regulatory networks and patterns hidden in heterogeneous human genome data across various biological conditions [1, 27], and discovered that lncRNAs possess a distinctive functional system that is distinct from that of proteins and is endogenous in the human genome independent of conditions [1]. For example, lncRNA transcription initiation and regulation are distinct from the mRNA-protein-based Pol system, and they are endogenous in the human genome across various physiological states measured by experimental designs deposited in the Sequence Read Archive (SRA) [29] database, which will be discussed in detail in this review.

This conceptual breakthrough in endogenous lncRNA functional systems lays a fundamental foundation for understanding the endogenous roles of lncRNAs in various physiological states such as tumorigenesis. As expected, another big data study revealed that lncRNAs serve as critical regulators in cancers and are endogenous in the cancerous region across all cancer types measured today by The Cancer Genome Atlas (TCGA) [30], whereas proteins only function under normal conditions [31].

This novel lncRNA functional system and its endogenous regulatory roles in the cancerous region have established a new conceptual framework for future functional studies of noncoding regions that occupy most of the human genome. Understanding these breakthrough concepts will help shorten the time frame required for hunting lncRNA functions. This review discusses the details of these discoveries and focuses on the conceptual breakthroughs. As this is not a comprehensive review, it does not contain basic information about lncRNAs. Readers interested in lncRNA comprehensive review of lncRNA function systems, please refer to recent excellent publications such as “Gene regulation by lncRNAs and its biological functions” by Statello et. al. [32], and “Mechanisms of lncRNA biogenesis as revealed by nascent transcriptomics” by Nojima and Proudfoot [14].

Before discussing biological concepts, it is necessary to briefly introduce the basic elements of big data studies. Massive data study relies on two key elements: (A) a massive, heterogeneous dataset that is large and heterogeneous enough to represent all conditions of a biological state and (B) a computational algorithm to generate unbiased results from a very large data set.

In contrast to conventional methods, which face challenges when dealing with heterogeneous data, the massive data approach welcomes heterogeneous data. Indeed, more heterogeneous data are better for big data analysis, in which more heterogeneous data help to generate more robust endogenous patterns. Therefore, the big data approach generally requires sufficiently large samples to represent all the conditions of a biological state [1]. For example, all human RNA sequencing (RNA-seq) samples from the SRA database that contains samples from almost all experimental conditions should be a substantial dataset representing all humans. In addition, all RNA-seq samples collected by TCGA from samples of the 36 most common types of cancer should constitute one massive dataset for all cancers. Other big data resources have been previously summarized [33].

On the other hand, big data studies require a computation algorithm capable of generating endogenous interactions from heterogeneous data after it has been collected. Although numerous algorithms and platforms have been applied to big biological data studies [34–37], such as artificial neural networks, support vector machines, and decision trees [35], these algorithms mostly perform classification. Conventional network inference algorithms such as C3NET and ARACNe-AP can infer gene interactions [38–41]. However, these conventional software packages suffer from high noise, which can contain more than 90% false positives [27, 41, 42]. In addition, machine learning with graphs [43, 44] has been widely applied to infer interaction networks, but it faces challenges in handling highly heterogeneous biological data with many more variables (genes) than observations (samples).

A novel software called Fast Inferring NETwork (FINET) [27] was developed to infer endogenous interaction networks from highly heterogeneous data. FINET infers any network quickly and accurately and infers endogenous regulatory interactions from highly heterogeneous biological data with > 94% precision as true positives/true positives + false positives. FINET speed and accuracy come from its implementation under fast Julia with stability selection, elastic-net machine learning, and parameter optimization algorithms. In addition to its first accuracy, FINET is user-friendly, with only a single command line to complete all computational processes, and it works in any OS system although FINET was developed under Linux. FINET is a critical tool for uncovering true interactions in big data. Understanding the algorithm helps in its wide applications. This review briefly describes the FINET algorithm and its application.

In a matrix with observations (bio-samples) as rows and variables (genes in biology) as columns, FINET [27] treats each gene as a target (set as y) and searches for its regulators from the remaining genes (set as X). The target-regulator interaction was inferred using an elastic net model [45].

The elastic net is arguably one of the most effective models for gene selection. Despite this, it is likely to produce more than 90% false positives when used alone to infer gene interactions in biology [27].

To reduce the false positives generated by the elastic net model, stability selection has been proposed [42]. Stability selection randomly splits the samples into two groups. When a target-regulator interaction is simultaneously selected in two groups, stability selection treats this interaction as true. Although this stability selection has been proven statistically, it could still contain more than 50% false positives in gene interaction inference because of heterogeneous data [27].

To improve the inferring precision and minimize false positive rates, FINET employs the following algorithm: FINET randomly splits the total samples into multiple groups, such as eight groups (m = 8), and independently infers gene regulatory interactions from each group using the elastic net (Figure 1). This process iterates n times. Interactions with high frequency during these m × n operations are treated as reliable interactions. Actually, the FINET algorithm filters condition-dependent interactions and retains conditionally independent interactions as endogenous ones.

Based on the algorithm described above, FINET can be widely applied to any type of big data. It has been applied to compute massive heterogeneous data, and its results have been validated [1, 28, 31]. For example, FINET has inferred endogenous regulatory lncRNA networks from all 265k human RNA-seq samples from the SRA database [1] and revealed endogenous lncRNAs from unannotated regions of the human genome [1, 31]. In addition, FINET unearthed an endogenous regulatory network for all cancers based on TCGA data [31]. Moreover, FINET has been applied to identify genome sequence motifs in evolutionary studies of the animal kingdom [28] and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus [46].

Together, the big data approach provides a promising way to minimize noise and understand the endogenous true picture of heterogeneous big data.

In contrast to conventional protein-dominated functional systems, noncoding RNA (ncRNA) functions are thought to be secondary in the human genome. However, a recent big data study revealed that ncRNAs, instead of proteins, drive animal lifespan evolution in the entire animal kingdom [28]. ncRNAs increase their content in animal genomes during evolution and coincide with trimming mitochondrial genome length, which is associated with lower energy consumption. Moreover, more active ncRNAs in the female reproductive system than their male counterparts account for why women outlive men [28]. These results indicate that ncRNAs are crucial functional genes in the two most important traits in humans, including reproduction and longevity. This discovery also emphasizes the biological significance of 98% of noncoding regions in the human genome and suggests that ncRNAs, rather than proteins, carry out most biological functions in the human genome. Understanding the functional system becomes critical for understanding human genome functions. However, the underlying ncRNA functional system remains largely unknown, and it is challenging to explore this complex ncRNA functional system in conventional studies.

A recent big data study was undertaken to capture the big picture of the lncRNA functional system in human genomes [1, 31], in which massive amounts of data were downloaded from the SRA [29] by searching human genomes and RNA sequences without filtering them out based on physiological conditions. This dataset contains all human RNA-seq data deposited in the SRA database, including 265,361 SRA samples under various experimental conditions, such as tissues, cell lines, and physiological conditions. These data are sufficiently comprehensive and heterogeneous to represent all types of conditions. Endogenous lncRNAs, including both annotated and unannotated lncRNAs, in this dataset reflect endogenous lncRNAs in heterogeneous human genomes. The functional characteristics of these endogenous lncRNAs represent the general endogenous traits of lncRNAs in all human genomes. In addition, the big picture derived from these endogenous lncRNAs represents the key biological principles of lncRNAs in all human genomes. As discussed in the following sections, these biological principles form the basis for understanding the functional system of lncRNAs in the human genome.

Scientific publications regarding lncRNAs and cancers have dramatically increased in recent years, from 627 papers in 2014 to 5,658 papers in 2020 [before the coronavirus disease 2019 (COVID-19) pandemic, Figure 3]. This indicates that the functional role of lncRNAs in cancer has attracted the attention of researchers. However, these studies did not identify the endogenous roles of lncRNAs in all cancers.

All cancers generally result from abnormal genomes evolving into an endogenous regulatory regime that is distinct from that in normal human tissues [61, 62]. Understanding this endogenous regime provides deep insights into the fundamental mechanisms of all cancers, and toward developing a general strategy to combat all cancers. Conventional approaches have been heavily employed to study this regime and to identify endogenous regulators of the cancerous regime [19, 30]. These approaches employ genome sequences, functional genomics, and biochemistry or combinations; however, the complex nature of cancer genomes and heterogeneous cancer data make these approaches ineffective. One of the highly intense study fields employed genome sequencing to identify mutations conserved across all cancers as universal cancer drivers. Projects based on this hypothesis have identified thousands of mutations in both protein-coding regions and ncRNAs in a large number of patient DNA samples from various cancer types [19]; however, no single consensus mutation has been found across all cancer types. Most of these mutations are specific to individual patients. For example, KRAS proto-oncogene, GTPase (KRAS) is one of the most mutated genes in lung cancer, but no single KRAS mutation is present in more than 40% of patients with lung cancer [19]. KRAS mutations are present in less than 2% of cancer types [19]. This indicates that the conventional strategy for identifying conserved mutations is unlikely to identify universal cancer drivers that endogenously regulate all cancer types.

Moreover, conventional biological approaches, such as gene knockout, usually cause transcript compensation [63] and alter whole-genome activation, leading to seriously biased gene regulation. Conventional computational approaches, such as regulatory network inference, usually suffer from high noise with a low accuracy of < 50% [27, 41, 42] when computing heterogeneously complex genome data. Taken together, these results suggest that the current conventional approach faces challenges when revealing endogenous cancerous mechanisms across cancer types.

Uncovering a systemic regulatory network that is endogenous to all cancers provides a foundation for understanding this regulatory regime. A recent big data study utilized FINET [27] to infer an endogenous regulatory network of annotated genes from massive heterogeneous cancer data, including all 11,574 RNA-seq samples and 36 cancer types measured using TCGA. This network discovery has advanced our knowledge of endogenous regulators that modulate this regime, leading to a conceptual breakthrough in cancer biology, as discussed in detail below.

With more functions of lncRNAs found in cancers, their clinical applications in cancers have increased linearly every year (Table S1). In 2022, there were six lncRNA clinical applications, and a total of 28 clinical applications were available in the clinicaltrials.gov database. These applications include drug targeting, diagnostic testing, and screening. Interestingly, exosomal lncRNAs have been used in cancer detection. This parallels our recent report showing that ncRNAs and an artificial intelligence (AI) neural network can detect all cancers with 96% of AUC.

Why the human genome contains 98% noncoding regions remains a mystery, but a recent big data study has dramatically advanced the understanding of its biological significance [28]. ncRNAs work with mitochondria in a low-energy fashion and extend animal lifespan during evolution; they also serve as crucial players in the human reproductive system [28]. Lifespan and reproducibility are the two most important traits in the evolution of any species. Without strong reproduction and long longevity, the species disappears in a long evolutionary process. Humans evolutionarily gain an increasing number of ncRNAs in their genomes to live longer and produce more generations. Assuming that these 98% noncoding regions were converted into protein-coding regions, humans might only survive a week as yeasts, which contain fewer noncoding regions and live only about a week. Therefore, 98% of these noncoding regions are functionally critical for humans. Consistently, recent big data studies have also unearthed ncRNAs as endogenous regulators for all cancers, whereas proteins only work endogenously under normal physiological states [31]. This suggests that under abnormal conditions, human ncRNAs are functionally more important than proteins. Humans live mostly in abnormal environments, such as various diseases, microbial infections, and variable environmental changes. These abnormal factors force humans to gain 98% of noncoding regions that are efficiently transcribed as ncRNAs that perform functions to cope with certain physiological states. Comprehending these ncRNA functions will be a key task in human genome research in the near future.

Comprehending ncRNA functions is challenging for conventional scientists, but big data scientists have taken advantage of the massive amount of available data and have revealed the big picture of human lncRNA functional systems [1]. In contrast to the conventional notion that lncRNAs are the secondary components in protein-based functional systems, the breakthrough discovery emphasized that lncRNAs have a distinctive functional system that is different from protein-based systems, and this system is endogenous to all human genomes and is independent of any condition [1]. This conceptual update of the endogenous lncRNA functional system has established a foundation for comprehending human genome function. For example, in this lncRNA system, lncRNAs have a unique transcription initiation system, although their initiation factors remain elusive. Future research using emerging big data will accelerate the discovery of lncRNA initiation factors. This will open a new avenue for understanding the transcriptional mechanisms of the dark regions in the human genome.

Since ncRNAs function as critical players under abnormal conditions and lncRNAs have their own functional system, it is not surprising for big data studies to uncover ncRNAs, instead of proteins as thought, as endogenous rulers for all cancers [31]. This breaks through the concept of fundamental drivers of all cancers. Future research based on this concept will help to elucidate the universal machinery and mechanisms of all cancers. The application of this novel concept will lead to the creation of a universal strategy for the diagnosis and treatment of all cancers. In particular, the biological functions of the abundant pseudogenes that dominate the cancer regime remain elusive. Elucidating the biological significance of these pseudogenes will be critical in cancer research and clinical applications.

Big data studies in biology are in their infancy, but they will evolve rapidly as emerging big data become available. Future research on big data should focus on both specific topics and big-picture analyses of a broad range of massive data to make conceptual breakthroughs in the general fundamental principles of biology. Embedded computational algorithms that integrate AI with emerging data requirements must be developed. These novel computational algorithms will advance the understanding of the fundamental principles of biology and lead to a revolutionary scientific discovery era.