Limits...
NNScore: a neural-network-based scoring function for the characterization of protein-ligand complexes.

Durrant JD, McCammon JA - J Chem Inf Model (2010)

Bottom Line: Although useful for in silico ligand identification, these scoring functions generally give many false positives and negatives; indeed, a properly trained human being can often assess ligand potency by visual inspection with greater accuracy.Computer-aided drug design depends on fast and accurate scoring functions to aid in the identification of small-molecule ligands.The scoring function presented here, used either on its own or in conjunction with other more traditional functions, could prove useful in future drug-discovery efforts.

View Article: PubMed Central - PubMed

Affiliation: Department of Chemistry & Biochemistry, NSF Center for Theoretical Biological Physics, National Biomedical Computation Resource, Howard Hughes Medical Institute, University of California San Diego, La Jolla, California 92093, USA. jdurrant@ucsd.edu

ABSTRACT
As high-throughput biochemical screens are both expensive and labor intensive, researchers in academia and industry are turning increasingly to virtual-screening methodologies. Virtual screening relies on scoring functions to quickly assess ligand potency. Although useful for in silico ligand identification, these scoring functions generally give many false positives and negatives; indeed, a properly trained human being can often assess ligand potency by visual inspection with greater accuracy. Given the success of the human mind at protein-ligand complex characterization, we present here a scoring function based on a neural network, a computational model that attempts to simulate, albeit inadequately, the microscopic organization of the brain. Computer-aided drug design depends on fast and accurate scoring functions to aid in the identification of small-molecule ligands. The scoring function presented here, used either on its own or in conjunction with other more traditional functions, could prove useful in future drug-discovery efforts.

Show MeSH
Average score (N) over 24 networks as a function of the experimentally measured Kd value. To facilitate visualization, the data were ordered by log10(Kd) value. Moving averages of both the log10(Kd) values and the associated N values were calculated over 100 points. This data-averaged function (shown in black) crosses the x axis at 25 μM [log10(25 × 10−6) = −4.60, shown as a dotted line]. Individual, unaveraged data points are shown in gray.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2964041&req=5

fig3: Average score (N) over 24 networks as a function of the experimentally measured Kd value. To facilitate visualization, the data were ordered by log10(Kd) value. Moving averages of both the log10(Kd) values and the associated N values were calculated over 100 points. This data-averaged function (shown in black) crosses the x axis at 25 μM [log10(25 × 10−6) = −4.60, shown as a dotted line]. Individual, unaveraged data points are shown in gray.

Mentions: To assess how the NNScore (N) varied according to the experimentally measured Kd values, we considered the scores of the 2710 characterized protein−ligand complexes with known Kd values described above. To facilitate visualization, the data were ordered by the log10(Kd) value. Moving averages of both the log10(Kd) values and the associated N values were calculated over 100 points and are plotted in Figure 3. Unaveraged data points are shown as gray circles. It is interesting to note that the data-averaged function crosses the x axis at roughly 25 μM [log10(25 × 10−6) = −4.60], as expected. So remarkable is this result that one again wonders whether the networks were overtrained; however, as Figure 2 demonstrates, these networks were consistently able to predict the binding of ligands to which they had never been exposed, suggesting the development of a genuine inductive bias.


NNScore: a neural-network-based scoring function for the characterization of protein-ligand complexes.

Durrant JD, McCammon JA - J Chem Inf Model (2010)

Average score (N) over 24 networks as a function of the experimentally measured Kd value. To facilitate visualization, the data were ordered by log10(Kd) value. Moving averages of both the log10(Kd) values and the associated N values were calculated over 100 points. This data-averaged function (shown in black) crosses the x axis at 25 μM [log10(25 × 10−6) = −4.60, shown as a dotted line]. Individual, unaveraged data points are shown in gray.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2964041&req=5

fig3: Average score (N) over 24 networks as a function of the experimentally measured Kd value. To facilitate visualization, the data were ordered by log10(Kd) value. Moving averages of both the log10(Kd) values and the associated N values were calculated over 100 points. This data-averaged function (shown in black) crosses the x axis at 25 μM [log10(25 × 10−6) = −4.60, shown as a dotted line]. Individual, unaveraged data points are shown in gray.
Mentions: To assess how the NNScore (N) varied according to the experimentally measured Kd values, we considered the scores of the 2710 characterized protein−ligand complexes with known Kd values described above. To facilitate visualization, the data were ordered by the log10(Kd) value. Moving averages of both the log10(Kd) values and the associated N values were calculated over 100 points and are plotted in Figure 3. Unaveraged data points are shown as gray circles. It is interesting to note that the data-averaged function crosses the x axis at roughly 25 μM [log10(25 × 10−6) = −4.60], as expected. So remarkable is this result that one again wonders whether the networks were overtrained; however, as Figure 2 demonstrates, these networks were consistently able to predict the binding of ligands to which they had never been exposed, suggesting the development of a genuine inductive bias.

Bottom Line: Although useful for in silico ligand identification, these scoring functions generally give many false positives and negatives; indeed, a properly trained human being can often assess ligand potency by visual inspection with greater accuracy.Computer-aided drug design depends on fast and accurate scoring functions to aid in the identification of small-molecule ligands.The scoring function presented here, used either on its own or in conjunction with other more traditional functions, could prove useful in future drug-discovery efforts.

View Article: PubMed Central - PubMed

Affiliation: Department of Chemistry & Biochemistry, NSF Center for Theoretical Biological Physics, National Biomedical Computation Resource, Howard Hughes Medical Institute, University of California San Diego, La Jolla, California 92093, USA. jdurrant@ucsd.edu

ABSTRACT
As high-throughput biochemical screens are both expensive and labor intensive, researchers in academia and industry are turning increasingly to virtual-screening methodologies. Virtual screening relies on scoring functions to quickly assess ligand potency. Although useful for in silico ligand identification, these scoring functions generally give many false positives and negatives; indeed, a properly trained human being can often assess ligand potency by visual inspection with greater accuracy. Given the success of the human mind at protein-ligand complex characterization, we present here a scoring function based on a neural network, a computational model that attempts to simulate, albeit inadequately, the microscopic organization of the brain. Computer-aided drug design depends on fast and accurate scoring functions to aid in the identification of small-molecule ligands. The scoring function presented here, used either on its own or in conjunction with other more traditional functions, could prove useful in future drug-discovery efforts.

Show MeSH