Model of Dry Lab

The modeling of our project consists of modeling of each part of the project and modeling of the project as a whole.

We modeled for the three parts of our project separately: wet lab, Tracer (hardware) and auto tester (software). In the part-by part modeling, we didn’t only model the performance of each part, but also the design of each part.

Wet Lab

Although we have designed and characterized three probes that can be induced-fluorescence, in order to extent the design of aptamer to the more target sequence and further optimize the design to achieve a higher ON/OFF ratio (the ratio of signal given when target exist over when target is absent), we would like to identified if there is any parameter(s) that allows us to built an automated system that provide a way for rational design of aptamer instead of testing randomly designed aptamers. To achieve that, we have tried to compare among the more well-functioning aptamers with those that does not function well.

Taking in an account in the assumption that the aptamer can achieve induce-fluoresce with a model that:

The truncated minispanich is destabilized in the aptamer so it would not be able to form the G-quaduplex that account to for the fluorescent
The destablized truncated minispanich aptamer can bind with target RNA to re-stabilize the structure,

As such, we define a “good” aptamer as those designs that:

Have low auto-fluorescent
Inducible by target

while, to path the way toward rational design of aptamers, the read out that we used from wet lab data are:

fluorescent count when only the RNA aptamer exists
given that the truncated minispanich is destabilized in the aptamer, the fold change in the fluorescent count of the experimental set which target is added, over which target not added

Part 1. Whether the truncated minispanich is destabilized in the aptamer (auto-fluoresce v.s induce-fluoresce)

Our designs are derived from a truncated minispanich (P1-a4-b5), which is produced by one base pair reduction (from 5 base pairs to 4) in stem a of the original fluoresing minispanich (P1-a5-b5) as described by Wei et al [1]. Therefore, we expect that a “well-destabilized” aptamer probe (and hence does not auto-fluoresce) are less likely to have strong interaction in the remaining 4 base pairs in stem a, which refers to the nucleotide 14-62, 15-61, 16-60 and 17-59 in our aptamer probe. Therefore, we built a scoring equation by multiple linear regression by plotting the mean fluorescent count of the aptamers designed (in the absence of target) from our wet lab experiment (N=17) against the binding probability of nucleotide 14-62, 15-61, 16-60 and 17-59 calculated by CentroidFold [2] and obtained the best fit (smallest R2) scoring equation as:

f(Score A)= -65075374(binding probability of nucleotide 14-62)+706797383(binding probability of nucleotide 15-61)-1617284819(binding probability of nucleotide 16-60)+27305386(binding probability of nucleotide 17-59)

High Score A suggests the aptamer design is more likely to be auto fluoresing. From the diagram above, the scoring equation shows distinct score for one of the aptamer probes showing strong auto-fluorescent but missed out another one. Also, this proposed scoring still has to be tested with more extra data with designs with high score and low score respectively. Yet, it can still be useful in screening out some of the aptamer designs that is likely to be auto-fluorescing (having high Score A) if it still stands after testing.

Part 2. Whether destabilized minispanich can be re-stabilized by target RNA(inducible v.s. uninducible)

Only considering the destabilizing on the minispanich is insufficient, as a probe that has to be induce-fluoresce is not necessarily inducible. Therefore, we would also like to have another score that allows us to identify those design that are more likely to be inducible, considering the ON/OFF ratio as the read out in our wet lab experiment that tells whether the probe is inducible.

Considering that the formation of heterodimer between the aptamer and target is in equilibrium, while the free energy (Delta G value) is related to thermodynamic stability, we selected the minimal free energy (MFE) of Aptamer-Target Heterodimer and Target-Target Homodimer, and the value of delta G for heterodimer binding to be plotted against the ON/OFF ratio to construct a best fit (smallest R2) equation using multiple linear regression. Also, we include frequency of the MFE structure in the ensemble, which may be related with the structural stability of the Aptamer-Target Heterodimer, in the multiple linear regression as a candidate that is related with the structural dynamics of the assembled heterodimer. The best fit equation to our data set (N=15) is found to be:

f(Score I)=3.71(MFE of aptamer monomer)+3.37(MFE of target monomer)-3.42(MFE of aptamer-target homodimer)+3.61(Delta G for heterodimer binding)+5.24(The frequency of the MFE structure in the ensemble)

The Score I shows positive correlation with the ON/OFF ratio, that those aptamer design having high Score I are more likely to be inducible. However, there are point that defect from the predicted trend and the R2 value only reaches 0.27.

This is not very surprising as a limitation of the regression plot only consider the variable included in the equation, while in the “turn-on” event, we only includes representative for molecular dynamics between aptamer and target, as well as representative for structural dynamics (The frequency of the MFE structure in the ensemble), but there is no representative for the molecular dynamics that account for the binding event of the heterodimer with DFHBI due to difficulties in predicting the interaction between an RNA and a small molecule (which is not an RNA).

Summary

Although the R2 value for both scoring methods are not high, they are still far from random. Therefore, it is reasonable to say that they have the potential to assist a rational design for minispanich aptamers that is more likely to be induced-fluoresce and inducible, and allows automated screening of aptamer designs along a long RNA sequence in the future by:

generate all possible aptamer design by sliding window of 22 nucleotide on a long RNA sequence that determines the variable domain of the minispanich aptamer probe
eliminate those are likely to be auto-fluoresing
select those are likely to be inducible

Reference:

W. Q. Ong, Y. R. Citron, S. Sekine, and B. Huang, “Live Cell imaging of endogenous mRNA Using RNA-Based fluorescence ‘turn-on’ probe,” ACS Chem. Biol., vol. 12, no. 1, pp. 200–205, 2017.
K. Sato, M. Hamada, K. Asai, and T. Mituyama, “CentroidFold: a web server for RNA secondary structure prediction.” Nucleic Acids Research. vol. 37, Web Server issue W277-W280, 2009.
L. Ivo, and Hofacker, “Vienna RNA secondary structure server” Nucleic Acids Research, vol. 31, no. 13, pp. 3429–3431, 2003.

Tracer modeling

For tracer we modeled the machanical design and the performance.

Mechanical design: All the 3D models of the mechanical designs of Tracer and its previous version can be found at our hardware page

Details of Hardware

Performance: The modeling of the performance of Tracer is composed of two major parts, accuracy and precision.

Accuracy: We use plate reader at our lab as the standard for accuracy analysis. The mobile phone we use to collect signal is iPhone 6S,
1. Method: We use GFP dilution as our sample for accuracy testing.
2. Steps:
  1. Prepare GFP dilution with different concentration.
  2. Analyse the pictures using matlab (See the appendix for detailed step of fluorescent level analysis using matlab)
  3. Put the dilutions in plate reader and check the fluorescent level
  4. Repeat the previous few steps and take the average value
3. Results:
4. Analysis:
  1. As can be seen from the graph the accuracy of Tracer is quite high compared to the results obtained from plate reader.
  2. However, at the low concentration region it becomes hard for tracer to distinguish between different concentrations (The same problem can be observed in the next part about precision). The reasons might be the following:
    1. The limitation of picture resolution due to the mobile phone camera
    2. The surface of the tube will also reflect some light. As a result, as can be seen from the following picture, most of the signal comes from the surface of the tube when the concentration is low. This also indicates that intensity may not be enough to distinguish positive and negative signals, and distribution is also a important parameter. This also accounts for the reason why we are using machine learning to process the signal image.
  3. Put the dilutions in plate reader and check the fluorescent level
  4. Repeat the previous few steps and take the average value
Precision:
1. Method: Prepare GFP dilution with concentration very close to each other. Here we managed to make the fluorescent intensity level of the dilution concentrated around the level of our RNA probe.
2. Steps:
  1. Prepare GFP dilution.
  2. Analyse the pictures using matlab
  3. Put the dilutions in plate reader and check the fluorescent level
  4. Repeat the previous few steps and take the average value
3. Results:
4. Analysis:
  1. As can be seen from the overall trend of the graph, Tracer can detect the difference of fluorescent around
  2. But similar to the above analysis for accuracy at low intensity range both plate reader and Tracer are not performing very well.

Team:Hong Kong-CUHK/model dry

Model of Dry Lab

Wet Lab

Tracer modeling