Final Presentation - December 13, 2002

MASKER estimated SES estimations were compared to analytically calculated values using MSMS by Bystroff.
• The version of MSMS used for this study has a known bug that gives the wrong values at certain distances in the 3-atom case (Bystroff and M. Sanner, personal communication).

• MSMS incorrectly shows an over-estimate and then a discontinuity at the distance where the reentrant surface goes to zero. The correct distance profile, and the MASKER profile, has the reentrant surface disappearing smoothly with increasing distance, as the self-intersecting part dominates.

• The toroidal and reentrant surfaces match the MSMS calculations over most of the trajectory but MASKER surfaces have small estimation errors.

• MASKER has greater computational overhead, since it must read a library of masks and mask data before beginning.

• The memory requirements of MASKER are also greater, again because of the library of masks.

• Bystroff believes that these binary operations will in the near future be encoded in special hardware. He states that additional speed-up would also be possible by initially computing the "reduced surface", and by using a neighbor-tree data structure for the atoms.

• According to Bystroff, the runtime [of Masker] is a polynomial with an exponent of approximately 1.42 and the process is dominated by the masking operations which consumes about 73% of the CPU time.

• The plot of the times that MASKER took to compute the molecular surface versus the number of atoms in the molecule fits best with the graph of polynomial with an exponent of approximately 1.64 (for proteins), 1.13 (for DNAs), 1.22 (for RNAs).

• The results clearly show consistency which is absent in some other algorithms.

• Draftpaper

Next slide