Biologic Institute

One of Biologic Institute’s research projects goes public this week, with a paper in PLoS ONE describing a new model for studying protein evolution [1] and a project at SourceForge.net providing the software to implement it [2] (both free of charge).

Okay—it’s not free beer, but here’s why we think it’s an exciting development anyway.

The challenge Darwinism currently faces is about broad explanatory principles—what kinds of things can be produced by undirected causes and what kinds of things can’t (even in billions of years). So, while the main goal is to explain the things of biology, it should be possible to advance that goal by examining things that are like the things of biology in some important way.

Computer models may prove very useful here.

There’s a catch, though. While it’s easy to capture some aspects of life in a computer model, it’s impossible to capture all of them. Who is to say, then, that one captures the key aspects of life and another doesn’t?

From our perspective, the aspect of life that seems to defy Darwinian explanation is its accomplishment of high-level tasks through the coordinated accomplishment of sub-tasks that occupy a succession of lower layers. Each layer requires the solution of its own set of design problems, all of these solutions serving the demands of the top-level solution. That will sound familiar to engineers, but this hierarchical way of tackling hard problems was first used in life.

And it has to be conceded that life employs it far more masterfully than humans do. To make our technological things, humans use large factories with supply chains that meet ISO 9001 standards, staffed with skilled people who work in clean rooms that have to be shut down regularly for maintenance. Life, on the other hand, is about things with on-board, self-maintaining factories that make the next batch of things—somehow. And to do this they require only a modest supply of air, dirt, rain, and sunshine.

Can such things evolve their way into existence? Computer models have shown that evolution of some kind is possible, but none have shown that the undirected invention of things of this kind is possible.

Take Avida, for example. [3] As a system for studying the evolution of artificial organisms, Avida is perhaps the most widely touted as having demonstrated that complex functional features can arise in a Darwinian way. [4] But do the evolved features really look life-like in their hierarchical complexity?

In two important respects they don’t.

In the first place, although Avida’s digital organisms solve problems, they do so in a quiz-taking sense rather than an engineering sense. Engineers solve real-world problems. Quiz takers solve artificial problems designed (by a teacher) to be readily solvable with the allotted resources. There’s a big difference between the two.

“Name a tool used by auto mechanics” is the kind of problem you might encounter on a quiz. “Fix my car” is the kind you would encounter as a repair mechanic. “Design a self-maintaining car” is the kind you would hope not to encounter as an engineer. “Design a self-maintaining car that makes more self-maintaining cars” is the kind you would rightly think impossible.

And yet, we have skittering, slithering, buzzing and blooming all around us umpteen million distinct and rather striking solutions to this impossible problem. Avida doesn’t touch that.

In the end, Avida tells us this: that computing devices supplied with lists of instructions that direct them to copy those instructions (with a possibility of introducing new instructions from a preexisting meaningful set) can, in an environment consisting of a master device that runs each new list on a computing device, produce additional instructions that cause their computing device to perform basic logic operations, like NOT, AND, OR, and EQU.

That might help an Avida organism to pass a quiz, but don’t expect to see these critters on the payroll of a software company any time soon.

This points to the other important disparity. Life builds things of stunning complexity from very simple building blocks, whereas the Avida inventions seem to be more like the reverse of this.

For example, cells build molecular machines (like the energy transducing ATPase described previously [5]) out of simple amino acids. But, as history proves, those amino acids can be studied at great length without anyone anticipating that machines can be built from them. Evidently there is genius in the design of molecular machines that is nowhere to be found in the amino acids themselves.

Contrast this with the Avida inventions. There even the most complex thing invented, the bit-wise equality function (EQU), seems simple in comparison to the supplied building blocks, which are machine-aware instructions like: “Test if two registers contain equal values”, or “Move a head by a fixed amount stored in a register”, or “Allocate memory for an offspring.” [6]

Most revealing, though, is the fact that Avida organisms, with all their front-loaded design complexity, fail to evolve EQU if the quiz grader doesn’t give credit for even simpler inventions. [4] In other words, Avida actually shows that Darwinian evolution goes nowhere unless certain rather fussy conditions are met. And when they are, it doesn’t go anywhere that should impress an engineer.

This is intriguing, to say the least. But other systems need to be examined.

Our Stylus software has something significant to offer here. Rather than emphasizing algorithmic replication, the way artificial life simulations like Avida do, Stylus emphasizes a life-like genetic structure and life-like causal connections between genes and their functions.

To achieve this, it depends on a real analogy. Like the structures of life, the structures of language are used to solve real problems at a high level. And the high level solutions in both worlds depend on a succession of solutions at lower levels.

In life, body plans serve the needs of particular modes of life, organs serve the needs of particular body plans, tissues serve the needs of particular organs, cells serve the needs of particular tissues, protein functions serve the needs of particular cells, protein structures serve the needs of particular protein functions, protein sequences serve the needs of particular structures, and genes serve the needs of these particular protein sequence requirements.

In a similarly hierarchical way, texts of various kinds serve the needs of particular communication objectives, sections serve the needs of particular texts, paragraphs serve the needs of particular sections, sentences serve the needs of particular paragraphs, phrases serve the needs of particular sentences, and words serve the needs of particular phrases.

What about letters serving the needs of words? Well, the problem with letter-based texts is that they are only sequences, whereas structures figure prominently in the functions of proteins. Protein sequences must form functional three-dimensional structures in order to work, whereas alphabetic sequences function directly as sequences.

But not all written languages are alphabetic. Chinese writing, in particular, employs structural characters that are analogous in some interesting ways to protein structures. Like folded proteins, these written characters perform the low level functions from which higher functions can be achieved.

Stylus builds on this analogy by using a life-like genetic code to specify simple building blocks (twenty vectors, analogous to the twenty amino acids) that in turn build Chinese characters. In this way, gene sequences looking very much like biological sequences encode vector chains with two-dimensional shapes. If those chains have the right geometry, by conforming to the shape of a character, they provide basic semantic function. And the functional hierarchy builds up from there.

The result is an artificial genetic system where genes encode basic functions by means of appropriate structures, and genomes encode higher functions that employ these basic functions. So, if it can be written in Chinese, it can be encoded in a genome and represented in working form by a proteome.

Cool, huh?

The big question, of course, is whether Darwinian evolution can do anything interesting in a system like this. In view of its similarity to life, the answer would be hard to ignore either way.

And the truth is—we don’t know the answer. Yet.

[1] http://www.plosone.org/doi/pone.0002246

[2] http://sourceforge.net/projects/biologicstylus/

[3] http://sourceforge.net/projects/avida

[4] http://www.ncbi.nlm.nih.gov/pubmed/12736677

[5] http://biologicinstitute.org/2008/04/03/perspectives/

[6] http://alice.cme.msu.edu/development/documentation/cpu_tour.html

rss — permalink