Limits...
Molecular counting by photobleaching in protein complexes with many subunits: best practices and application to the cellulose synthesis complex.

Chen Y, Deffenbaugh NC, Anderson CT, Hancock WO - Mol. Biol. Cell (2014)

Bottom Line: The step detection algorithms account for changes in signal variance due to changing numbers of fluorophores, and the subsequent analysis avoids common problems associated with fitting multiple Gaussian functions to binned histogram data.The analysis indicates that at least 10 GFP-AtCESA3 molecules can exist in each particle.These procedures can be applied to photobleaching data for any protein complex with large numbers of fluorescently tagged subunits, providing a new analytical tool with which to probe complex composition and stoichiometry.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Huck Institutes of the Life Sciences, University Park, PA 16802 Interdisciplinary Graduate Degree Program in Cell and Developmental Biology, Huck Institutes of the Life Sciences, University Park, PA 16802.

Show MeSH

Related in: MedlinePlus

Detecting steps in simulated stepping data. (A) Histograms of step sizes predicted by all step detection algorithms. The simulated data have uniform step sizes of 1 with 10% backward steps and SNR of 1. Real step sizes are calculated by comparing the means of plateau regions on either side of a step. The mode at +1 represents forward steps, and the mode at −1 represents backward steps. The four algorithms detect unitary forward and backward steps but also have modes centered at +2, corresponding to twice the single step size and representing missed steps. (B) Sensitivity plots for the four algorithms. The missed steps corresponding to the lower sensitivity of Bdetector2 can be seen in A by the population centered at +2 step size. (C) Precision plots for the four algorithms. Bdetector1 had problems with overfitting, resulting in lower precision and a number of steps between 0 and 1 in A.
© Copyright Policy - creative-commons
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4230622&req=5

Figure 3: Detecting steps in simulated stepping data. (A) Histograms of step sizes predicted by all step detection algorithms. The simulated data have uniform step sizes of 1 with 10% backward steps and SNR of 1. Real step sizes are calculated by comparing the means of plateau regions on either side of a step. The mode at +1 represents forward steps, and the mode at −1 represents backward steps. The four algorithms detect unitary forward and backward steps but also have modes centered at +2, corresponding to twice the single step size and representing missed steps. (B) Sensitivity plots for the four algorithms. The missed steps corresponding to the lower sensitivity of Bdetector2 can be seen in A by the population centered at +2 step size. (C) Precision plots for the four algorithms. Bdetector1 had problems with overfitting, resulting in lower precision and a number of steps between 0 and 1 in A.

Mentions: To validate their performance, we first tested the step detection algorithms on simulated stepping data having SNR values from 0.4 to 5 (Figure 3). The step times were sampled from an exponential distribution with an expected value of 100 time points/plateau, with 90% of steps being a unit step increase and 10% being a unit step decrease. At high SNR values, the mean predicted step size was close to the actual value, but with diminishing SNR, an additional peak corresponding to twice the unitary step size emerged (Figure 3A and Supplemental Figure S2). We defined two metrics, sensitivity and precision, to assess the performance of the algorithms. Sensitivity is defined as the proportion of the true steps that are identified by the step detection algorithm. Precision is defined as the proportion of identified steps that are true steps (see Materials and Methods). Overfitting will lead to high sensitivity and low precision (false positives), whereas underfitting results in high precision but low sensitivity (missed events). With SNR values >2, all four algorithms performed well and had both high-sensitivity and high-precision values (Figure 3, B and C). Reasonable predictions were obtained at SNR values between 1 and 2, but sensitivity and precision both fell sharply for SNR values <1. The BIC-based algorithms displayed a tradeoff between sensitivity and precision, with Bdetector1 (constant variance) having higher sensitivity and Bdetector2 (unequal variance) having higher precision (Figure 3, B and C, blue and green plots). In contrast, for the two-sample t test methods, Tdetector1 (assumed constant variance) and Tedector2 (assumed unequal variance) performed similarly (Figure 3, B and C, red and black plots).


Molecular counting by photobleaching in protein complexes with many subunits: best practices and application to the cellulose synthesis complex.

Chen Y, Deffenbaugh NC, Anderson CT, Hancock WO - Mol. Biol. Cell (2014)

Detecting steps in simulated stepping data. (A) Histograms of step sizes predicted by all step detection algorithms. The simulated data have uniform step sizes of 1 with 10% backward steps and SNR of 1. Real step sizes are calculated by comparing the means of plateau regions on either side of a step. The mode at +1 represents forward steps, and the mode at −1 represents backward steps. The four algorithms detect unitary forward and backward steps but also have modes centered at +2, corresponding to twice the single step size and representing missed steps. (B) Sensitivity plots for the four algorithms. The missed steps corresponding to the lower sensitivity of Bdetector2 can be seen in A by the population centered at +2 step size. (C) Precision plots for the four algorithms. Bdetector1 had problems with overfitting, resulting in lower precision and a number of steps between 0 and 1 in A.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4230622&req=5

Figure 3: Detecting steps in simulated stepping data. (A) Histograms of step sizes predicted by all step detection algorithms. The simulated data have uniform step sizes of 1 with 10% backward steps and SNR of 1. Real step sizes are calculated by comparing the means of plateau regions on either side of a step. The mode at +1 represents forward steps, and the mode at −1 represents backward steps. The four algorithms detect unitary forward and backward steps but also have modes centered at +2, corresponding to twice the single step size and representing missed steps. (B) Sensitivity plots for the four algorithms. The missed steps corresponding to the lower sensitivity of Bdetector2 can be seen in A by the population centered at +2 step size. (C) Precision plots for the four algorithms. Bdetector1 had problems with overfitting, resulting in lower precision and a number of steps between 0 and 1 in A.
Mentions: To validate their performance, we first tested the step detection algorithms on simulated stepping data having SNR values from 0.4 to 5 (Figure 3). The step times were sampled from an exponential distribution with an expected value of 100 time points/plateau, with 90% of steps being a unit step increase and 10% being a unit step decrease. At high SNR values, the mean predicted step size was close to the actual value, but with diminishing SNR, an additional peak corresponding to twice the unitary step size emerged (Figure 3A and Supplemental Figure S2). We defined two metrics, sensitivity and precision, to assess the performance of the algorithms. Sensitivity is defined as the proportion of the true steps that are identified by the step detection algorithm. Precision is defined as the proportion of identified steps that are true steps (see Materials and Methods). Overfitting will lead to high sensitivity and low precision (false positives), whereas underfitting results in high precision but low sensitivity (missed events). With SNR values >2, all four algorithms performed well and had both high-sensitivity and high-precision values (Figure 3, B and C). Reasonable predictions were obtained at SNR values between 1 and 2, but sensitivity and precision both fell sharply for SNR values <1. The BIC-based algorithms displayed a tradeoff between sensitivity and precision, with Bdetector1 (constant variance) having higher sensitivity and Bdetector2 (unequal variance) having higher precision (Figure 3, B and C, blue and green plots). In contrast, for the two-sample t test methods, Tdetector1 (assumed constant variance) and Tedector2 (assumed unequal variance) performed similarly (Figure 3, B and C, red and black plots).

Bottom Line: The step detection algorithms account for changes in signal variance due to changing numbers of fluorophores, and the subsequent analysis avoids common problems associated with fitting multiple Gaussian functions to binned histogram data.The analysis indicates that at least 10 GFP-AtCESA3 molecules can exist in each particle.These procedures can be applied to photobleaching data for any protein complex with large numbers of fluorescently tagged subunits, providing a new analytical tool with which to probe complex composition and stoichiometry.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Huck Institutes of the Life Sciences, University Park, PA 16802 Interdisciplinary Graduate Degree Program in Cell and Developmental Biology, Huck Institutes of the Life Sciences, University Park, PA 16802.

Show MeSH
Related in: MedlinePlus