Introduction

Protein interaction is widespread in all live units, and its importance is self-evident. In our design, red light can stimulate the combination of N-hrluc-PhyB and C-hrluc-PIF3 and then produce blue light to guide microalgae to move(Fig.1).Therefore, whether our design is reasonable and the experiment successes depends on the degree of combination of N-hrluc-PhyB and C-hrluc-PIF3. Based on chemical thermodynamic formula, the relationship between binding rate and concentration and affinity constant is established in two steps. In addition, as a commonly used light-controlled polymeric protein actuator, this model can not only guide our own experiment, but also provide a quantitative model for the field of optogenetics.

Design

The molecular reaction in the solution can be divided into two steps:the first step is red light stimulation of PhyB and PIF3 binding, the second step is that the binding of PhyB and PIF3 facilitates the binding of two parts(NHrluc、Chrluc) of the Sea kidney luciferase. The molecular reactions carried out in our model and solution are also divided into two steps.

1.1 The first part: only the binding of PhyB protein and PIF3 protein is considered.

In order to find out the effect of the initial concentration of PhyB protein and PIF3 protein on the binding rate of the two proteins, we considered the binding of only PhyB and PIF3 proteins, and through of two kinds of proteins .Then,we get a specific equation between the initial concentration of protein and the binding rate of the two proteins.

1.1.1  Parameter

Symbol	Meaning
A	Concentration of PhyB
B	Concentration of PIF3
Ka	Affinity constant
Kd	Dissociation constant
η	Initial concentration of PhyB
A0	Initial concentration of PhyB
B0	Initial concentration of PIF3
AB	Concentration of PhyB-PIF3

1.1.2  Hypothesis:

(1) After the binding of the two proteins, the solution contains three substances: protein PhyB, protein PIF3 and the combination of the two. That is, the material is conserved and there is no protein loss before and after the reaction.
(2)   Except for the initial protein concentration, other factors have the same effect on the binding of the two proteins.

1.1.3   Modeling

Considering the binding of PhyB protein and PIF3 protein in solution, there are/p>

If the free concentration of PhyB is A, the free concentration of PIF3 is B, and the concentration of PIF3 is AB, the binding constant is:

The dissociation constant is

Then the expression of the binding rate is

Available jointly by (2) (3) (4):

Because we assume that the initial concentration of PhyB protein is higher than that of PIF3 protein, that is,

Available jointly by (5)(6):

The specific equation between the initial concentration of the two proteins and the binding rate of the two proteins was obtained.Available from (7)

When the light with the wavelength of 650 nm and a frequency of 20 Hz, we know Kd=500nM [1]（fig.1）：

The main results are as follows: (1) similar to the barrel effect, the binding rate depends on the low concentration of protein, and it can be estimated from the diagram that the minimum threshold of the concentration of the two proteins is mol /L at the same time.

(2)If the combination rate of A and B is not considered under red light (dark), then

Under the excitation of red light,the value is drastically reduced due to the change of in the configuration of the protein ,only when,The combination rate in the dark can be ignored.

1.2 The second part: the combination of N-hrluc and C-hrluc.

Under the action of "optical switch", the distance between N-hrluc and C-hrluc becomes closer, which is equivalent to increasing the concentration. The effect of "optical switch" on protein binding is explored by using the definition of equivalent concentration to replace the "optical switch".

1.2.1   Parameter

Symbol	Meaning
Bi,i=1,2	Represent N-hrluc and C-hrluc protein parts respectively.
Ai,i=1,2	Represent PhyB and PIF3 respectively.
d	Distance between B1 and B2

1.2.2  Hypothesis:

(1) The addition of optical switch has no effect on the conformational and properties of the original protein.
(2)   After the optical switch binds, the original protein is pulled closer, that is, the distance is smaller.

1.2.3   Modeling:

As shown in the following figure, optical switches are added to N-hrluc and C-hrluc proteins .Under the irradiation of light,the optical switches begin to combine.However, due to the conformational structure of B1 and B2 does not change, so the affinity constant is unchanged, but the distance of B1and B2 is close, which is equivalent to the increase of concentration.

The equivalent concentration of a particle can be obtained as follows:

The volume can be defined by the spacing between proteins, that is,

So:

According to the size of the protein, in the least ideal case, the distance between N-hrluc and C-hrluc is 16 nm. Assuming that the spacing is 15 nm, 10 nm and 5 nm, the equivalent concentrations are mol/L、 mol/L、mol/L。

As long as the affinity constant of the two parts of fluorescein enzyme is less than ,it can be considered to be 100% binding.

1.3 Conclusion:

our system is mainly limited by optical switches, so the Chlorella vulgaris need to be able to express enough PhyB and PIF3.

1.4 Sensitivity analysis:

If we do not conduct plasmid, add sea kidney luciferase and sea kidney fluorescein to the culture medium directly, and using the red light stimulates the movement of microalgae, and determines the threshold of sea kidney luciferase, which is what we want to verify through experiments.

1.5 Prediction:

If mol/L 's sea kidney luciferase can make microalgae move, then according to our model, it is assumed that under the same light condition, the value under red light is much smaller than that under dark, and the affinity constant between the two parts of sea kidney fluorescein enzyme is less than , equivalent to simultaneous expression in microalgae mol/L N-hrluc-PhyB and C-hrluc-PIF of mol/L).

Introduction

After subtilisin-modified ribonuclease was split in 1958，multifarious proteins was split successfully certificated by experiments. Our experiments can split Renilla luciferase and control the activity of it with optogenetics methods. The two parts of the split Renilla luciferase need to be self-assembled under the guidance of the optical switch. Different splitting sites will produce two different parts, and the assembly effect will be different. The splitting site is the result of experiments by the predecessors, which is time-consuming and laborious, so whether can we predict better sites without experimental methods? We have searched a large number of previous experimental results and found that the resolution site is closely related to the secondary structure of the protein, and the resolution site is generally located in the loop region of the protein. In addition, the location of the split seems to be related to factors such as Sequence conservation, Length of coil regions, and Center distance. Based on these rules we have summarized, we established the Protein-split Model to predict better protein splitting positions. We chose nanoLuc protein in the iGEM library as an example. The combination of the two nanoLuc parts obtained by the split position of our model is better than their original one, which is a good way to verify our model.

Design

2.1 Three factors

We conclude that the protein-split sites may be related to three factors, Secondary structure (factor 1), Length of coil region(factor 2), Center distance(factor 3)(fig.1), and found 65 Splitting sites (table.1) by experiment.

2.2 BP neural network

This is a typical BP neural network binary classification model. If we can select a suitable set of weights{w1, w2} and correspond to any set of samples in the learning sample, the inputs (I1, I2, I3) = (Factor1, Factor2, Factor3), output(O1, O2)=(1,0). Then the protein-split sites selection is actually solved. For any protein, we only need to download its secondary structure from the PDB database, each Amino acid residue corresponds to a set of factors. As long as the factors are input into the neural network, a Label will be output: T:1:[1, 0] ; F: 0: [0, 1] (fig. 2). We use the training of the sample to find a set of appropriate weights {w1, w2} to complete our model, and do the sensitivity and specification analysis.

Algorithm

See connected protein-split_supplement1.

Database

Input see connected protein-split_supplement2,Output see connected protein-split_supplement2.

Result

Peural network parameter

w1

1	2	3
-0.0306	0.233	0.5924
0.6037	-0.6431	-0.5429
1.3603	-0.6622	0.5412
0.6023	-0.3277	-0.5616

b1

0.7124	-0.464	-0.9178	-0.5768

w2

1	2
0.3485	-0.1632
-0.2581	0.1562
-0.8925	0.9408
-0.2699	-0.1891

b2

0.7532	0.1274

Performance Evaluation Criteria

Our model is highly sensitive, but have low specificity which produces high ratios of false positives, meaning that many bad protein-split sites are recognized as good protein-split sites. However, in general, a good cleavage site of a protein with hundreds or thousands of amino acid residues is much less than a bad cleavage site, so the low specificity of this model is inevitable, we only need to ensure that very few good sites are not mistaken for a bad place. This model will provide several predicted good sites during the actual instructional experiment, as long as there are true good sites. Because our model has a very low false negative, this model is actually very good in practical applications.

Experimental Verification