Why is artificial intelligence not ready to deal with our complex genomes?
The DNA inside our cells is not just a set of instructions, but something much more dynamic. Like scientists, AI models don't yet have the data to properly decipher.
In fact, the human genome is less a script than a puzzle that gets harder the closer we look. Knowing the whole sequence—the order of the roughly 3 billion chemical blocks of our DNA, deduced almost entirely by the International Human Genome Project between 1990 and 2003—hasn't helped much. That research showed that barely 2% of the human genome is made up of actual genes, that is, sequences that code for DNA information.
It is now clear that understanding the human genome no longer depends on what each gene does. The deeper and more difficult question is how these genes are used or regulated, a question that seems to involve some, perhaps much, of the rest of the genome. By turning on and off a set of genes, different types of cells in our body can be created from the same material. Cells also regulate their genes moment by moment in response to a constant stream of signals from their neighbors and surroundings. But the processes that govern gene regulation are so complex that some biologists wonder if a complete understanding—of how the genome actually works—is within our feeble minds.
Some are counting on outsourcing analytics to artificial intelligence. Genomic "fundamental models" such as Google DeepMind's Evo 2, Genos, and AlphaGenome are trained on vast amounts of genomic data, which biologists use to predict how differences in DNA sequence affect biological processes and ultimately traits (including disease risk) of an entire organism. These algorithms are not concerned with complex monitoring issues. All of this apparently feeds into the "training" of the algorithm, through which it infers correlations from what it already knows.
This approach is probably useful, but for those who want a true understanding of how the genome works and ultimately life itself, a computational black box will never suffice. And perhaps more to the point, the genome may not lend itself to the kind of straightforward input-output approach that such AI models ultimately assume.
This is because the genome has no blueprint or algorithm. It is something else
The traditional view holds that a small part of our DNA contains the code to make the protein molecules that regulate the chemistry of our cells. Each instruction for a protein is stored in a corresponding gene—we have about 20,000 of them—and gene sequences can range in length from a few dozen to nearly 3 million DNA "letters" (representing molecules called nucleotides).
Making a protein from its gene is a two-step process. First, the DNA is read, letter by letter, by an enzyme called polymerase, which makes a copy of that code into a related molecule called messenger RNA (mRNA). This is called transcription. The mRNA is then read by a piece of molecular machinery called a ribosome, which makes a protein—a process called translation. The proteins made by the ribosomes are then moved to carry out their functions in building and maintaining the organism.
This picture is more or less true. But Adelman said that "genes are probably not the most interesting part of the genome."
What matters more is how our genes, many of which we share with simpler organisms, are regulated—turned on and off. Which protein a cell needs changes over time and by cell type: muscle, brain, skin, etc. How the genes that code for those proteins are regulated depends on some parts of the genome that don't code for proteins.
Biologists have known about gene regulation and the involvement of "non-coding" DNA since the 1960s. But for many years, most of what they understood came from studies of simple organisms like bacteria, whose principles are generally simple. However, it gradually became clear that in complex eukaryotic organisms like us, gene regulation is much more complex, involving overlapping monitoring and control systems, each with its own complexities.
Transcription is initiated by proteins called transcription factors, which act as managers of gene regulation. These proteins attach to sections of DNA (usually near the target gene) and use the polymerase enzyme to make a copy of the mRNA. In bacteria, transcription factors are more like keys that match locks to unique binding sites on DNA. But this is not how they work in complex organisms. In us, logic is more difficult than analyzing transcription factors.
For one thing, our transcription factors do not show strong preferences for specific DNA binding sites. Additionally, they tend to work in pairs or groups. And a given transcription factor may have different effects in different contexts, such as activating gene transcription in one cell type but repressing it in another, depending on which other transcription factors are around.
In bacteria, according to Adelman, regulation has an "OR" logic whereby a particular signal turns a gene on or off: it's either this or that. But in the human genome, the logic is more like what computer scientists call "AND." Many signals are integrated to reach a regulatory decision: this and that as well as the other thing. In this case, the setting can be more responsive to context differences, and the setting buttons are configurable rather than just on or off. "That's part of the beauty" of the complexity of our regulations, Adelman said.
When they interact with the genome, transcription factors bind to stretches of DNA called enhancers—which create a puzzle of their own.
Enhancers recruit sites for transcription factors and are thought to have a decisive effect on transcription: they send a "go" signal to a waiting polymerase to make an mRNA copy of the DNA sequence. It sounds simple enough, but mapping enhancers to their corresponding genes is not so simple. Our genome may be hundreds of thousands, millions of enhancers, which means that we outnumber our genes. Each gene may be affected by many enhancers, and each enhancer may affect several genes.
Wendy Bickmore, a genome biologist at the University of Edinburgh, said: "It's a shame that 25 years after the Human Genome Project, we don't know where all the enhancers are in the genome, let alone what they do when they do and which genes they control.
Biologists know that most enhancers do not respond to a transcription factor. Their activation "requires a cocktail," Bickmore said. "That's what gives [the enhancer] its extraordinary property — because it's only in a specific cell at a specific time that you have the right combination of factors to bind and activate that enhancer."
Some enhancers are, as you might expect, close to the genes they regulate, or even on the DNA within a gene. But others are far from the gene—perhaps millions of nucleotides away, with many more genes in between.
The presence of such so-called "distal" enhancers "seems mysterious," Bickmore said. "How do you get that information from there to here, to the gene that needs to be activated? That's a largely unanswered question."
One of the answers comes in the form of a loop.
Distal enhancers are brought to the gene they regulate on large loops of DNA or, more precisely, chromatin, the combination of DNA and its packaging proteins that unfolds as if from a ball of wool. These loops are created by a protein motor called cohesin, which moves the DNA strand up and down and extrudes it as needed.
Once cohesin forms a loop to bring elements together, what then? It was once thought that they would then stick together or assemble into a molecular machine, but this is not the case. Instead, the components appear to form a loose but dense bubble in which they interact weakly, ephemerally, and uncertainly—a kind of committee sometimes called condensation.
These transcription hubs are highly fluid and vary from cell to cell. "There's some ring extrusion going on here, in the next cell it might be here, and everything is spinning at an incredible speed," Bickmore says. Even if the cells are conceptually identical—say, two skin cells—exactly what the gene regulatory machinery is doing at any given moment is never the same in any two of them.
Chromatin loops are just one reason why the transcription of a gene depends on the shape and structure of the surrounding chromatin.
Certain enzymes are involved in packaging and repackaging of chromatin, thereby controlling transcription. In other words, what is important is not only the information encoded in DNA, but also how it physically and dynamically exists in space. "We no longer think of the genome as a linear piece of DNA code," Bickmore said. "It's a very exciting change to think about this very dynamic 3D folding that's completely inherent in regulation."
One aspect of this three-dimensional organization is the clustering of chromatin segments into compartments called topologically associated domains (TADs). In a TAD, genes seem to be regulated together: they turn on or off in groups. Such groups keep a set of genes on or off to shape and function in different cell types. Cohesin is also involved in assembling chromatin to form TADs—a dynamic process in which chromatin is constantly rearranged in our cells.
Chromatin shape can also be affected by chemical changes called epigenetic marks: small molecules attached to DNA packaging proteins called histones or attached directly to DNA. Some of these epigenetic changes can alter the electrical charges on histones, which alters how proteins attract or repel each other and thus change chromatin packing. Epigenetic modifications of chromatin are like annotations of the DNA script, changing its meaning in a given context. When cells divide, epigenetic annotations are also copied.
How and when markers are added and changed, and what each marker means for gene activity, are complex questions with no simple answers. Some researchers speak of an "epigenetic code" that governs this aspect of gene regulation, but it's unclear whether there really is anything this systematic.
All of these and other processes can determine whether a gene is transcribed into mRNA. But there are many more layers of regulation that determine which ayasps mRNA is translated into the corresponding protein—and which protein is made.
This post-transcriptional regulation is often controlled by RNA molecules that are said to be non-coding. These short-lived molecules, such as mRNA, are not templates for proteins, but they have their own functions. While mRNA is produced from protein-coding regions of DNA (so-called "coding genes"), non-coding RNAs are transcribed from other regions of DNA, now generally described as non-coding genes. These non-coding RNAs are versatile and assume different roles in a cell. Researchers are learning more about what they can do every day, and many, if not most, of them appear to be involved in gene regulation.
For example, small non-coding RNAs called microRNAs can silence mRNAs before they are translated into proteins. They do this by directing special enzymes to a specific mRNA to break it down or chemically modify it. microRNAs do not do this alone, but not unlike transcription factors, they act in combination, in groups, and in a relatively promiscuous manner: a given microRNA may regulate many mRNAs, and a particular mRNA may be regulated by many microRNAs.
Why make an mRNA only to prevent it from being translated into a protein? This type of post-transcriptional regulation is like having another checkpoint: does the cell really need this protein? MicroRNAs can be mobilized to allow cells to regulate gene expression depending on the immediate context. In this way, the functioning of the genome is less like a program of inevitable progress and more like an adaptive and responsive process.
Another post-transcriptional complication is that mRNAs are translated into protein only after reorganization. Fresh from transcription, mRNA contains sequences that encode bits of protein called exons, as well as sequences that should not be translated and need to be spliced out, called introns. (Strictly speaking, this pre-edited RNA is called pre-mRNA.) The job of editing introns and splicing exons together is done by a molecular assembly called the spliceosome, which is made up of several proteins along with various non-coding RNAs.
The spliceosome can also be tissue-sensitive, so that it may splice pre-mRNA to encode one protein in one cell type and a slightly different protein in another. Sometimes these different protein "isoforms" can have very different roles. For example, transcription factors often bind alternately in this way, and their isoforms can assume different regulatory functions—for example, some may activate gene expression, while others repress it.
All told, these and other regulatory mechanisms show that the genome is far from an automatic program running in the background to build us and keep us alive. Our cells actually make complex decisions about how to use their genes—both the information they carry and the structure they assume.
Therefore, cells need to assemble a relatively loose and ambiguous committee of components, such as transcription factors and enhancers, to initiate transcription, which also depends on how the chromatin strand is formed and shaped at that moment. Then there are more layers of decision-making and action between the mRNA and the final functional protein.
Also remember that all actors—from transcription factors to non-coding RNAs—are themselves generated from the genome in some kind of context-dependent process. This makes the genome more like a recursive, self-referential system that computer scientist Douglas Hofstadter called a "strange loop." It acts on itself, according to its history—which determines, for example, its chromatin composition and epigenetic marks—and according to messages inside and outside the cell. So there is no plan.
And because of that, it is not easy to understand. "If I were God, I wouldn't have designed it that way," Bickmore said. "But here we are!"
Why is gene regulation so complex in animals like us? One potential answer is that evolution does not have the foresight to design with clear efficiency and logic, but simply manipulates what was already available. Maybe so, but eukaryotic gene regulation isn't just a messed-up version of what happens in bacteria. It has different principles and surely there is a reason for them.
Bickmore speculates that the complexity of regulation and organization of the genome may have been the only means of creating complexity in the organism. For example, organisms with different types of tissues and diverse lifestyles needed more control over whether genes are on or off in a particular cell. One thing this required was more and more non-coding regulatory sequences in DNA. But then all of them could not get close to Gene himself.