In evolutionary informatics, at least two paramaters are needed for evolutionary algorithms.
Heritability and Selection.
Heritability implies the following:
- “Parents” give rise to “offspring”.
- Traits from “parents” are passed on to “offspring”.
- Each “offspring” from a “parent” signifies a new generation.
- Variation between generations may or may not occur.
Selection implies that certain traits that are not on a fitness landscape will not be selected.
Let’s look at Autodock as an example and how it relates to evolutionary informatics. Autodock employs a genetic evolutionary algorithm in order to try and predict the orientation of a ligand within a protein.
The ligand is the heritable structure. (A ligand is any structure that binds to a protein, e.g. a therapeutic molecule)
The protein is the fitness landscape.
The genetic evolutionary algorithm provides the variation and selection parameters.
Consider the following diagram:
Figure 1: A) Basic lay out of memetic algorithms. A population of individuals is randomly seeded with regard to fitness (initialized). The individuals are randomly mutated and their fitness is measured. Individuals with optimal fitness are further mutated until convergence of a local optima is reached. The process is carried out for the entire initialized population. The global optima is selected from the various local optima. B) Fitness landscape with local optima (A, B and D) and a global optima (C). In a memetic algorithm, the initial population of individual are randomly seeded and can be viewed as any of the arrows indicated in the figure.
A few important aspects from the figure:
- Fitness depends on the phenotype.
- Fitness (in the case of Autodock) is the capability of the ligand phenotype to bind and stay bound to the protein.
- The parameters for succesful binding are many. For Autodock, the following are included:
Van der Waals interactions
Hydrogen bond interactions
Torsional free energy
If certain parameters (above) are not on a fitness landscape for a certain ligand phenotype such as the absence of hydrogen bonds at a particular area of the protein, such a trait will not aid in ligand binding for a particular ligand with hydrogen bonds. Therefore,hydrogen bonding (as a trait) will not be on the fitness landscpe and is thus not a selectable trait.
Autodock uses a Solis & Wets search algorithm to probe the fitness landscape of a particular protein. (See figure below)
The surface of a protein is where the binding of the ligand will occur, thus 3-dimensionally, the fitness landscape would look something like this:
Rapamycin ligand bound to the mTOR protein.
So how does the algorithm find the local optima within proteins?
With autodock, a population of individuals (ligands) are randomly placed within the receptor. The conformation ligand-protein interactions are measured for each individual and is then followed by a conformational “mutation” (See image below).
The binding energy for each conformation “mutation” is measured until a local optima for a specific population of individuals is reached. The binding energy of the local optima of each population is measured, and the global optima is the population of individuals that have the best binding energy (See below).
If the evolutionary algorithm is well designed, the conformation of the global optima will correspond to the experimentally determined crystallographic pose. The Root Means Squared Deviation (RMSD) of a docked ligand compared the to the crystallographic pose is generally used as a good indicator. A RMSD value less than 2 is considered a success. In the case of the Autodock software, the global optima is supposed to correlate with the crystallographic pose (RMSD <2).
As an example, a ligand was docked into a protein with the following results.
Docked ligand positions and binding energies
As seen here, the global optima corresponded reasonably well to the crystallographic pose (RMSD<1.8 ), meaning the software sucessfully probed the fitness landscape of the protein to find the optimal solution.
Autodock is thus a nice example of how evolutionary informatics and evolutionary design principles can be applied to design optimal structures such as therapeutically relevant compounds/ligands.
Now let’s consider another example in nature and how heritability and selection is applied.
As an example, consider the following diagram.
A fitness landscape (From here)
Again, a few important aspects from the figure:
- Fitness depends on the phenotype
- Fitness in this case is the capability of the phenotype to reproduce (self-replicate)
- The parameters for succesful self-replication are many. A few examples:
A) Fast replicators (e.g. bacteria)
B) Intelligent replicators (e.g. monkeys)
C) Cooperative replicators (e.g. ants)
D) A combination of the above (e.g. humans)
E) Population dynamics
F) And others…etc.
Therefore, if certain parameters are not on a fitness landscape for a certain phenotype (such as the capacity to construct a car, such a trait will not be selected in the next generation if the population of phenotypes consist of bacteria.)
The aim of this thread is to:
- Discuss evolutionary dynamics and fitness landscapes and how it is related to nature and other evolutionary algorithms.
- See if there are any parallels between the two examples of how evolutionary informatics are applied in molecular biology.
- How evolutionary dynamics and evolutionary design principles can be applied to real world problems.