log inhelp

 | 
View
 

List of Digital Evolution Research Papers

Page history last edited by Wolfgang Banzhaf 2 years, 6 months ago

Books: 

 

 

Evolution of Complexity:

 

 

 

The Simplicity Bias of Evolution

 

 

 

Neuroevolution (Evolution of neural networks, artificial brains, intelligence)

 

Introductory papers (from the column in the Artificial Life Journal: "The Evolutionary Path to Sentient Machines")

 

 

Flow of information:

 

  • C Bohm, D Kirkpatrick, V Cao, C Adami. Information Fragmentation, Encryption and Information Flow in Complex Biological Networks. Entropy 24 (2022) 735.
  • A Hintze and C Adami, Cryptic information transfer in differently-trained recurrent neural networks. Proc. 7th International Conference on Soft Computing & Machine Intelligence (ISCMI), p. 115-120.
  • A Hintze and C Adami, Neuroevolution gives rise to more focused information transfer compared to backpropagation in recurrent neural networks. Neural Computing and Applications (2022) 022-08125. 
  • A Tehrani-Saleh and C Adami,  Can Transfer Entropy Infer Information Flow in Neuronal Circuits for Cognitive Processing? Entropy 22 (2020) 385.
  • D Kirkpatrick, V Cao, and C Bohm, Tracking data flow in digital brains exposes coincidental encryption. ALIFE 2022: The 2022 Conference on Artificial Life, 114 (2022). p. 114. 

 

Neural Architecture

 

 

Evolution of Navigation, Foraging, Learning, and other Behaviors

 

 

Evolutionary Machine Learning

 

 

Methods to Improve Evolutionary Computation

 

 

Evolution of Recombination and Sexual Selection

 

 

 

Neutral Evolution

 

 

 

Genetic Architecture

 

 

Genetic Improvement

 

 

 

Evolution of Cooperation and Altruism

 

Evolution of Coordination / Group Selection

 

 

Evolutionary History and Phylogenetic Reconstruction

 

 

Speciation, Ecology, and Coevolution

 

 

Genetic Languages

 

 

Tempo and Mode of Evolution

 

 

Mutations and Evolution of Robustness

 

 

Evolution of Evolvability 

 

 

Abundance Distributions

 

 

Origin of Life 

 

 

Long-Term and Open-ended Evolution

 

 

 

Evolutionary Applications in Engineering

 

 

 

Virus Evolution

 

From the 2022 Class: A collection of papers on Virus Evolution and on Predicting Evolution:

 

Synopsis: In this paper, we study the coevolution of a viral quasi-species and an immune quasi-species. The theory predicts that both the viral quasispecies and the immune quasispecies evolve towards their optimal mutation rate: one mutation per unit of time that it takes the immune system to adapt to a new viral epitope. In other words, according to this theory the virus receptor binding domain changes just when the immune system has figured out the latest change.  It is not clear whether we can observe this dynamics in SARS-Cov2.

Synopsis: This paper introduces the antibody-escape calculator. For each of 31 known SARS-Cov2 antibodies (both natural and designed) the calculator aggregates data from Deep Mutational Scanning (DMS) experiments that measure binding of antibody to virus (several different variants) with all possible single-site mutations in a 200 amino acid stretch of the receptor binding domain (RBD). Based on this data, you can predict the likelihood of anti-body escape for any combination of mutations you enter, for any combination of antibodies. Based on this calculator, Jesse was able to predict that Omicron will escape binding almost completely (meaning that any previous immunity to Delta, as well as double-vaccination, are not protective).  That prediction turned out to be accurate. The drawback is that because they only tested every single mutation, interactions between mutations (epistasis) are not accurately reproduced.

Synopsis: This paper studies single point-mutations on SARS-Cov2 lineages and identifies fitness-changing mutations. "Mutational fitness" is the percentage change in growth rate (technically, it is the log of R_child/R_parent, where R is the growth rate of a lineage). There is an inverse correlation between growth rate and mutational fitness, because the higher the R of a lineage, the smaller the increase in growth rate (diminishing returns). This paper also performs a simulation of lineage evolution (see Section: "Simulation of lineages") which could be used as a starting point for lineage evolution simulations. 

Synopsis: This paper introduces a tool that can link sequence to function using an information-theoretic approach. The framework goes beyond standard approaches (such as regression and Potts-model) and takes into account correlations between sites that goes beyond the pairwise approximation. It is successfully applied to avida sequence data (26 symbols), but should be readily applicable to amino acid or nucleotide sequences

Synopsis: This paper uses Deep Learning to predict infectivity and antibody-binding resistance from sequence data. A possible project would be to use the tool developed by Nitash in the paper just above this one to see if it can do better than Deep Learning. 

Synopsis: This paper documents the evolution of SARS-Cov2 in a single immune-compromised patient in South Africa over a period of six months. During that time, the virus evolved from the ancestral strain to evade antibodies to a number of different strains (including Beta and Delta), to end up with a type that looks very similar to Omicron. 

This paper was the result of Jesse Bloom's project in a class very similar to this one that I taught at Caltech at the time. There are some really interesting conclusions in this paper that I do not remember, such as that high-stability proteins have a hard time evolving new functions. But this may be very different from evolving anti-function, that is not binding to a particular target ligand. As far as I know, nobody has done that experiment before. One thing to note: at the time Jesse was limited in the length of proteins to investigate by the computational power of the day. Eighteen years later, we can probably do a heckuva lot better!

Synopsis: Why do we never have to update the vaccine for measles (same vaccine since the 1960s) while those for influenza and Covid have to be constantly updated? This paper is a commentary by Jesse Bloom's group (on another research paper) arguing that the measles antibody targets numerous distinct viral epitopes, making it immune to evasion. This may also be tested in a lattice protein evolution simulation. 

Synopsis: This is a computational study of quasispecies dynamics in finite populations. The genotypes here are simulated RNA sequences that fold into secondary structures (kind of like the lattice protein model, but for RNA). "selection for mutational robustness" is another term for "survival of the flattest" dynamics, which is going hand-in-hand with quasispecies dynamics. In another words, if you have high mutation rate, then you can have quasispecies and selection for mutational robustness, which creates the opportunity for "survival of the flattest". In this paper, we show that this also works for finite populations. In particular, Fig. 6 shows that the consensus sequence in the population constantly changes. This does not happen unless you have a quasispecies. It would be great if we could document this in an actual population, but no single patient has its viral population sequenced to that extent. There is data about genotype frequencies within local regions (Aimer has found lots of data for Colorado, for example) but finding that the consensus sequence in a population changes is not the same thing has showing that the consensus sequence in a patient changes. 

 

Below are a selection of papers that make use of digital evolution, both with implicit and explicit fitness functions, both for traditional population genetics topics, as well as for more behavioral biology/evolution of intelligence topics.  Any of these papers can be chosen to be replicated and extended as part of your course project.  Feel free to also suggest other papers not listed here. Most papers have links that take you either to the journal/abstract, or directly download the pdf of the paper.

 

 

 

Comments (0)

You don't have permission to comment on this page.

Printable version