Some Paradoxes of Causation

What do we usually mean by the cause of a phenomenon?

Rules and/or facts by which some behavior can be accurately predicted.

For example, the three-dimensional folding shape of a protein is caused by

    1) The amino acid sequence. ("primary sequence")

    2) Forces of interaction between parts of this amino acid chain (including hydrogen bonding between different amino acids, hydrogen bonding with surrounding water, forced bending by prolines, bending encouraged by glycines, disulfide bonds, hydrophobic "bonds" between side chains of valine, leucine and isoleucine, etc.

    3) Sometimes heat shock proteins, or other chaperonins.

Computer programs have been written that predict folding shapes of actual proteins, based on their amino acid sequences. These programs work partly by calculating forces of attraction, like hydrogen bonds, but also depend heavily on noticing similarities to amino acid sequences of other proteins, for which the 3-D structure was already known. These computer programs predict protein structures correctly around 80% of the time; definitely nowhere near 100%. Why not?

Please notice that this is not the same as using X-ray diffraction, applied to pure crystals of a given protein, to calculate three-dimensional structures. X-ray diffraction always gives the correct structure, unless *The protein won't crystallize; *The protein folds into 2 or more different patterns, which mix together randomly; *Or if the protein can't be purified. (Or gets damaged by isolation, etc.). These issues are also worthy of thought.

Proteins whose function is to bind selectively to specific DNA sequences and thereby favor or inhibit gene transcription are called "transcription factors". Obviously, it is very important to be able to predict what DNA base sequence will bind to a transcription factor having a given amino acid sequence. Therefore, people have tried to write computer programs to predict base sequence binding to proteins having a given amino acid sequence. This is even more difficult than writing a program that predicts protein shape, because the program also has to predict the DNA sequence that the protein will bind to. Guess what percentage of the time programs solve this problem correctly. Would you believe "never"?

Other double-jointed prediction problems include binding of one protein to other proteins.

Chess playing computer programs work partly by calculating effects of all legal moves, for 2 or 3 or 4 generations of future moves (e.g. If I move my queen to that space, I can capture my opponent's rook, but what can he then do to me? Etc.) But chess playing computers also make much use of similarities, for example, to a game played in 1980, when Fisher did take the opponent's rook with his queen, with the result that...whatever. In other words, not just algorithms, but also stored memory.

Weather prediction computer programs are based largely on physics. This includes physical observations, like gradients of humidity and temperature; and also includes physical processes, like how much heat will be released by a given amount of condensation of rain. Not so many people realize that these computer programs are also based on historical "memorization" of previous days on which the spatial distribution of temperatures and humidities happened to be very similar to what they are today. If so, the weather tomorrow gets predicted to be the same as the weather was the day after that previous day.

Look at the possible future paths of Hurricane Nadine, now out in the middle of the Atlantic. These have unusually much divergence.

THE NEXT QUESTION is how do proteins (or weather) calculate their own future behavior? If super-computers can't always predict accurately the shape into which a protein will fold, then how do we expect the poor little protein to decide on some consistent shape?

PLEASE google "Levinthal's Paradox"

Read the Wikipedia article on this subject, which is very good. One good thing is that the "talk" page includes examples of different misunderstandings of this subtle and difficult topic. Somebody thought it was a hoax. Others claim it isn't really a paradox. One person thinks chaperonins are a way to explain it. One argues that the problem is avoided by synthesis of proteins one amino acid at a time. If that were the answer, then would denatured proteins ever re-fold back to the same original geometry? Another person confidently believes that proteins do not conduct random searches of alternative conformations. Yet another guesses that way back in 1969, when Levinthal published his famous "thought experiment" (In PNAS), people didn't yet know as much as they do now about protein folding. In case that seems likely to you, forgive me for mentioning that in 1968 I happened to have taken an excellent course on protein conformation taught by Prof. Frederick M. Richards. Try Googling his name. Everyone in that class was directly exposed to the what you might call the full blast of the leading expert on protein conformation, and I have kept up with the subject since then. Nothing has emerged that eliminates Levinthal's paradox.

Please be ready to discuss what you think might be the relationship between Levinthal's paradox and the difficulty of predicting protein folding.

Are these two problems opposite sides of the same coin?

What experiments might be able to find out?

Don't panic, because nobody knows what is the true answer.

The answer may be (I bet) related to prions, mad-cow disease, Alzheimer's disease.

Both problems (computer predictions and Levinthal's paradox) bring us face-to-face with very important unsolved problems, including whether over-confidence on "Genomics" is built on logical fallacies.

If you know the exact amino acid sequence of a protein, or of several related proteins, how much can you really predict?
Is this an inherent, fundamental problem; or can it be solved by improved methods, or concepts?

(And don't worry about quantum theory; we will just assume these are not "Copenhagen Interpretation" problems.)




back to index page