Prediction of mutations engineered by randomness in H5N1 hemagglutinins of influenza

[DreamSciTech]

Re: Prediction of mutations engineered by randomness in H5N1 hemagglutinins of influe

Dear All

Many thanks for your warm words and encouragements. Sorry for the delay in answering you because we were busy all the day, and until now we have time to see this website.

At first, let us talk several words regarding the logistic regression and neural network. As we wrote in our previous posts, assume that we have an equation ax + b = y for describing the cause-mutation relationship. We have said that y is the mutation, in fact, it is the occurrence/non-occurrence of mutation, which can be classified as unity and zero, a binary event. In such a case, on the left-hand side of equation we have continuous value, x, and on the right-hand side we have a binary event. This type of problem is the problem of classification, which can be solved using either logistic regression or neural network. At the beginning of our study on prediction of mutation, we used the logistic regression, which has an explicit form and works very fast. However, the logistic regression is not powerful enough, which is more suitable for concept-initiated study, and now we began to use the neural network, which is very time-consuming and does not have any explicit form. This is our current stage of studies.

With respect to the timing of mutation, we consider that the prediction of mutation should include (i) the prediction of mutation position, which we are trying to use the neural network to solve, (ii) the prediction of would-be-mutated amino acids at predicted positions, which we are trying to use the amino-acid mutating probability to solve, (iii) the timing of mutation, which we are trying to use the fast Fourier transform to solve, and (iv) the prediction of new function in mutant proteins.

The fast Fourier transform, in plain words, is a mathematic tool, which can find the periodicity in chaotic data, for example, we at first hope to find if there are periodicities in the figure in www.dreamscitech.com, whose periodicity is not easily to be seen in figure. When using the fast Fourier transform to treat the data from in figure, we can find that there are many periodicities. With the clear periodicities, we can time the mutations although there is still a long way to go.

Good weekend

Guang and Shaomin

[DreamSciTech]

Re: Prediction of mutations engineered by randomness in H5N1 hemagglutinins of influe

Hi, Laidback Al

We just realized that our explanation of timing of mutations is less understandable in our previous post, thus during the weekend we made two figures in our website, www.dreamscitech.com, to show the results of fast Fourier transform on the historical data of hemagglutinins, and how we stratify the historical data into several segments to time the mutations. Hopefully, these figures would be more understandable for timing of mutation.

Have a great week

Guang and Shaomin

[Laidback Al]

Re: Prediction of mutations engineered by randomness in H5N1 hemagglutinins of influe

Thank you for posting the chart. I will spend some time studying it as will others who read this thread.

[DreamSciTech]

Re: Prediction of mutations engineered by randomness in H5N1 hemagglutinins of influe

Dear All

Happy Thanksgiving Day! Now we made the data for constructing the figure of evolution of hemagglutinins downloadable. You are kindly invited to download the data at www.dreamscitech.com to conduct your analysis to time the possible outbreak of influenza.

All the best

Guang and Shaomin

[DreamSciTech]

Re: Prediction of mutations engineered by randomness in H5N1 hemagglutinins of influe

Merry Christmas and Happy New Year! Everyone


This week, we found that we have finished the calculations of H5N1 for 2006, and found that there are only 5 new H5N1 sequences in influenza database in 2007. After finished the calculation of these five new sequences, the total H5N1 sequences in 2007 reached 50.

However, an important difference appears between the five new H5N1 sequences and previous 45 ones, that is, the mean +/- standard deviation are 39.03 +/- 0.96 for five new sequences but 37.60 +/- 1.08 for the previous 45 H5N1 sequences, p = 0.007, Mann-Whitney test.

Suggestive here is that the curve began to change its direction slowly, and direct to upwards. If the historical observation holds on, say, the outbreak occurred after extreme point (see the red cycles in figure), then a bird flu related to H5N1 appears beyond horizon. Of course, there is an important difference in historical figure, where we calculated all subtypes of hemagglutinins, but now we only calculated H5N1 hemagglutinins recently.




Could the third pair be matched in the lower figure (the filled cycles are outbreaks)?

Guang and Shaomin

[Laidback Al]

Re: Prediction of mutations engineered by randomness in H5N1 hemagglutinins of influe

Guang and Shaomin, thanks for posting the graph and making the data available on your web site.

If I understand correctly the most recent five H5N1 samples as a group are statistically different than the previous 45 samples from 2007. Can you tell us which five samples these were and where they were collected?

The data for the past several years shows that the mean values fall outside the 95% confidence interval. Does this recent extreme deviation from the regression line have any significance?

If the historical observation holds on, say, the outbreak occurred after extreme point (see the red cycles in figure), then a bird flu related to H5N1 appears beyond horizon.

Does this quote mean that the potential for a pandemic increases when the trend line reaches an inflection point on the graph and the "predictable portion of randomness" increases.