EVE: An Artificial Intelligence at the Interface Between Evolutionary Medicine and Genomic Medicine » The Sankore Writings » SciLogs

Evolutionary medicine is a branch of human medicine in which doctors search for new therapeutic approaches to treat human diseases in animal metabolism. I would like to explain this using the example of a hereditary liver disease: people who suffer from Progressive familial intrahepatic cholestasis type 3 (PFIC3) suffer from extremely low levels of phospholipids in the bile, which is associated with an increased risk of gallstones and destruction of the bile ducts. PFIC3 begins in early childhood and, because of its severe course, often necessitates a liver transplant. PFIC3 is caused by mutations in the two Abcb4 inherited from the mother and father1-Genes that cause these genes to be inactive and the cells do not make Abcb4 protein. Only when the maternal and paternal Abcb4 genes are inactive does PFIC3 break out.

Example of a missense mutation: In the DNA sequence, one of the nucleotides (A = deoxyadenosine triphosphate) is replaced by another (C = deoxycytidine triphosphate). This results in an incorrect amino acid being incorporated into the protein sequence.

Orthologous genes and hereditary diseases

The Abcb4 gene codes for a protein that is produced in the liver and from there transports phospholipids to the bile. The Abcb4 gene is an orthologous gene. Orthologous genes are genes that occur in different organisms (eg human, mouse, fly, baker’s yeast, sunflower) and whose nucleotide sequence is very similar. In terms of phylogenetic history, a common ancestor is assumed in this case, from which the corresponding organisms evolved. Because of the high agreement2 in the nucleotide sequence, a similar function of the proteins they encode is assumed. The more closely related animal species are, the higher the similarity of the nucleotide sequences between two orthologous genes. The more closely related animal species are, the more orthologous genes they have in common. Because geneticists have been working for the past 24 years3 not only determined the sequence of the human genome, but also the genomes of many animals, many orthologous genes have already been discovered, which in e.g. Orthodisease be documented. OrthoDisease is a database of model organism genes that are orthologous to human genes that have mutations that result in inherited diseases. These animal genes are in English Orthologs for human diseases called.

See also  No stress or Alzheimer's, memory loss and high cholesterol could reveal the deficiency of this ignored mineral

The Abcb4 gene is inactive in guinea pigs and horses because large parts of the DNA sequence in this gene are missing. In addition, guinea pigs and horses have extremely low levels of phospholipids in their bile, just like people suffering from PFIC3. This raises the question of why these two mammalian species do not suffer from the liver damage typical of PFIC3, even though the gene is inactive. How does the liver stay healthy? To answer these questions, human medicine relies on the help of veterinary medicine and bioinformatics. Bioinformatics now features artificial intelligence EVE (Evolutionary Model of Variant Effect) to support them with a new tool.

Artificial neural networks and unsupervised machine learning

EVE is an artificial neural network that uses Unsupervised Machine Learning (UML). UML is not based on predefined parameters and rules, but includes adaptive learning. A classic example is learning from cat and dog images which are cats and dogs: in the UML the network is presented with a series of cat and dog images without being told which images are cats and which are dogs. It has to admit it itself. UML recognizes new patterns in data that have never been seen before, making this approach particularly suited to the task.

Schematic representation of an artificial neural network with several hidden layers (here a feedback model, since the information only propagates forward). An artificial neural network consists of nodes, also called neurons, which receive information from other neurons or from outside, change it and send it out as a result. This is done via three different layers, each of which can be assigned a type of neuron: those for the input (input layer), for the output (output layer) and the so-called hidden neurons (hidden layers) that are between the input layer and the output layer. .

Researchers Debora S. Marks (Harvard University Medical School, USA) and Yarin Gal (University of Oxford, England) and their team developed EVE to detect mutations in the orthologous genes of 140,000 animal species (including extinct species) to recognize and to classify. . The researchers used EVE to assess 36 million protein sequences and 3,219 disease-related genes in different animal species. EVE estimated the probability that each individual amino acid variant is either safe or harmful to humans.

See also  A plaque on the waterfront and a history in medicine: who is Dr. Lattmannn

To determine whether EVE makes accurate predictions, the researchers compared its results with five mutations in human genes known to be clinically important: various forms of cancer, various cancer syndromes and cardiac arrhythmias. EVE’s predictions overlapped with current descriptions of experimental data.

A notable advantage of EVE over current methods is that it assigns a continuous score rather than a binary score. Because even if gene variants are classified as causing disease, the way a mutation can affect the body is complex. “There’s a whole continuum of pathogenicity,” says Marks. “The continuous count is very important for predicting the level of pathogenicity. Does the mutation mean that I will get pain in my pinky toe or will I die tomorrow?” The researchers must B. take into account that an orthology of genes does not necessarily lead to an orthology of organs. For example, it cannot be ruled out that orthologous genes and thus also orthologous proteins are found in two completely different, non-orthologous tissues.

Another important aspect of EVE is the assignment of a prediction certainty score for each individual gene. For each genetic variant, EVE tells doctors how much they can trust the prediction. “We don’t just give medical professionals a number, we also give them the level of uncertainty that comes with it,” says Gal. “It’s something they can use in decision-making. EVE can say, ‘I think this variant belongs in that stack, but I’ve never seen variants like that, so take it with a grain of salt.

From Orthodisease to Genomic Medicine

Marks and Gal also take part in the Atlas of Variant Effects Alliance, a global research initiative that aims to create a comprehensive atlas of all possible human gene variants and their impact on protein function and organs. The goal of the efforts is to improve the diagnosis, prognosis and treatment of human diseases. Since Next generation sequencing4 Automated genome sequencing on a large scale has made it possible for researchers to more quickly understand how specific genetic changes affect a person’s health or disease. Genomic medicine uses sequence information for genetic diagnostics and clinical interpretation of individual genetic information.

See also  Worse infections as the disease progresses - Medicine/Therapy - Multiple Sclerosis News


1. Abcb4 is the English abbreviation for ATP binding cassette subfamily B member 4

2. If two genes match in the nucleotide sequence by more than 30% of their nucleotides in the sequence, then a cause other than common descent is considered unlikely.

3. In 1998 the genome of the roundworm Caenorhabditis elegans order. It was the first time that an animal’s genome had been sequenced. The first mammalian genome was that of humans in 2003.

4. An entire human genome can be sequenced in a day.

Further reading

1. Frazer, J., Notin, P., Mafalda, D., Gomez, A., Min, JK, Brock, K., Gal, Y., Marks, DS. (2021) Disease variant prediction with deep generative models from evolutionary data. Nature 599, 91–95.

2. Human genome now fully sequenced



Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Social Media

Most Popular

On Key

Related Posts