Free Energy Calculation
in MD Simulation
Basic Thermodynamics
Helmoholtz free energy
A = U TS ! " #
i
$
i
dA = %
re&
're&ersi(le) const $ * T+ e, '--./+ Mc0uarrie 1 Simon
2i((s free energy
2 = U ! 3* TS ! " #
i
$
i
= H TS ! " #
i
$
i
d2 = %
non3*
're&ersi(le) const $ 3 T+ e, '--.45+ Mc0uarrie 1 Simon
U internal energy 3 3ressure #
i
Chemical 6otential
* *olume T Tem6erature
S Entro6y dS = d,7T S = 8
B
ln 9
:m6lication of Free Energy
A ; B <
e,
= =A>7=B>
<
e,
= e?6 '@A2
B
7CT+
A2
B
= @CT ln <
e,
A2 = A2
B
! CT ln 0
A2 D B Unfa&ora(le
A2 = B
A2 E B Fa&ora(le
Statistical Mechanics
System can (e descri(ed using Hamiltonian
H'6
4
)6
-
).........6
$
) r
4
) r
-
).......r
$
+
Different ensem(le 'fi?ed system ,uantities+
@ Canonical ensem(le '$)*)T+
@ $3T ensem(le
Statistical Mechanics
Common Free Energy Ty6e
Sol&ation Free Energy 7 Transfer Free Energy
Binding Free energy
Confomational Free Energy
Calculating Free Energy
E?6erimentally
@ 6ro(a(ilities of finding the system at gi&en states
A2 = @CT ln '3
4
73
B
+
@ re&ersi(le %or8 of mo&ing the system (et%een
t%o states
Com6utationally
@ 6o(a(ility
@ re&ersi(le %or8 F more efficient
3ro(a(ility Method ' Brute Force +
Ergodic hy6othesis
time a&erage = ensem(le a&erage
A2 = @CT ln '3
4
73
B
+
:s this trueG :n general) $HI
@ sam6ling time is too short com6utationally
@ hard to sam6le e&ery state of the system
For system state of zero sam6ling) 3
A
= B F A2
B F A
= J
A=lim
t -
1
t
A( t ) dt=
1
M
M
A
i
= A
Alanine Di6e6tide
Challenge
Accurate calculations of a(solute free energy is
nearly im6ossi(le due to insufficient sam6ling in a
finite length and time scale simulation.
$eed different methods to estimate free energy.
Common method
@ Thermodynamical integration
@ Free energy 6ertur(ation
@ Um(rella sam6ling
@ 3otential of mean force
Thermodynamic Cycle
H ( x , p
x
, \)=H
0
( x , p
x
)+\ H
b
( x , p
x
)+(1\) H
a
( x , p
x
)
Thermodynamic :ntegration
Thermodynamic :ntegration
the &alue of EdA7dKD is accurately determined
for a num(er of intermediate &alues of K) the
total free energy is determined %ith numerical
integration methods (ased on these &alues
A'K+ is smooth enough and con&erged
:ntermediate &aluesL
fluctuation of MH7MK for each &alue of dA7dK
Free Energy 3ertur(ation
Error Estimation
Chris Chi6ot N $AMD mailing list
:n FE3) con&ergence may (e 6ro(ed (y monitoring the time@e&olution
of the ensem(le a&erage. This is) ho%e&er) a necessary) (ut not
sufficient condition for con&ergence) (ecause a66arent 6lateaus of
the ensem(le a&erage often conceal anomalous o&erla6 of the
density of states characterizing the initial and the final states. The
latter should (e the 8ey@criterion to ascertain the local con&ergence of
the simulation for those degrees of freedom that are effecti&ely
sam6led.
Statistical errors in FE3 calculations may (e estimated (y means of a
first@order e?6ansion of the free energy) %hich in&ol&es an estimation
of the sam6ling ratio of the latter of the calculation 'Straatsma) 4/O5+.
htt6L77%%%.cirm.uni&@rs.fr7&ideos7-BB57e?6oses7B-PQeBris7Chi6ot.6df
E?am6le
Mutate 3HE to H:S
3henylalanine Histidine
F to H S9$T dU
3henylalanine &s. Histidine
A2
3HE
= @-.RST 8cal7mol
A2
H:S
= @4.RTS 8cal7mol
H
U
$
CH C
CH
U
HH
H
From amino acid S9$T free energy 6rofile) %e can estimate A2
4
and A2
-
L
A2
4
= @-.RST 8cal7mol A2
-
= @4.RTS 8cal7mol
AA2 = A2
-
@ A2
4
= +0.991 kcal/mol +0.991 kcal/mol
From alchemical transformationL
A2
4
alch
= @-O.VO 8cal7mol A2
-
alch
= @-R.OO 8cal7mol
AA2W = A2
-
alch
@A2
4
alch
= +0.7 kcal/mol +0.7 kcal/mol
1
alch
2
alch 1 2 (aq)
G
(aq)
G G
1
alch 2
2
alch 1 (aq)
G
(aq)
G G G G SWNT - HIS HIS
G G G G SWNT - PHE PHE
2
2
alch
1
alch
1
=
+ = +
To (e continued............
Thermodynamic :ntegration
H ( p
1
, p
2
, ....... p
N
, r
1
, r
2
, .......... r
N
)
A A
a-b
=A((
b
)A((
a
)=
dA(()
d (
d (
dA(()
d (
=
H ( r , p)
(
exH (r , p) dr dp
exH (r , p) dr dp
A A
a-b
=
H ( r , p)
(
d (
H ( r , p)
(
= F
(
=constraint force
Ceaction Coordinate X
Constraint force L force re,uired
to constrain the system at a fi?
reaction coordinate X
E?am6le
Translocation of C5B into D33C (ilayer
Ceaction coordinate L Distance (et%een the
CHM of C5B and CHM of D33C mem(rane
SS constrained 6oints L B S- Y) 4 Y s6acing
Harmonic constraint force
8 = 4BB 8cal7mol7Y
-
Um(rella Sam6ling
Sam6le %ith um(rella 6otential
UZ'?+
Com6ute (iased 6ro(a(ility
3Z'?+
Estimate un(iased free energy
A ? = [8 ( )
B
T ln 3Z ? [ UZ ? F ( ) ( ) +
F is undetermined (ut irrele&ant
Com6le? surfaces ha&e multi6le (arriers
$eed to 8no% the free energy surface to 8no% an efficient
(ias
Harmonic (iasing function
Multi6le simulations
3ut the minimum of the (ias in a different 6lace for each
simulation 'sam6ling %indo%s+
Estimate 3Z'?+ for each simulation
Com(ine results from all simulations
From one simulation
A ? = [8 ( )
B
T ln 3Z ? [ UZ ? F ( ) ( ) +
F de6ends on UZ'?+
Different simulations ha&e different offsets
$ot o(&ious ho% to com(ine different %indo%s
9HAM
9eighted Histogram Analysis Method
Determines o6timal F &alues for com(ining
simulations
<umar) et al. \ Com6ut Chem) 4S) 4B44@4B-4)
4//-
2eneralizations
Multidimension reaction coordinates
Multi6le tem6eratures
$
sims
= num(er of simulations
n
i
'?+= num(er of counts in histogram (in associated %ith ?
U
(ias)i
) F
i
= (iasing 6otential and free energy shift from simulation i
3'?+ = (est estimate of un(iased 6ro(a(ility distri(ution
F
i
and 3'?+ are un8no%ns
Sol&e (y iteration to self consistency
Cunning US simulation
Choose the reaction coordinate
Choose the num(er of %indo%s and the (iasing
6otential
Cun the simulations
Com6ute time series for the &alue of the
reaction coordinate 'histograms+
A66ly the 9HAM e,uations
E?am6le
Translocation of C5B into D33C (ilayer
Ceaction coordinate L Distance (et%een the
CHM of C5B and CHM of D33C mem(rane
/ %indo%s L B S- Y) T Y s6acing
Harmonic constraint force
8 = B.V 8cal7mol7Y
-
Hther Methods
Steered MD
Ada6ti&e Biased Force Method 'ABF+
3arallel Tem6ering 'Ce6lica e?change+
@ o&ercome free energy (arriers
etc.
Ce6lica E?change
H&ercoming Free Energy Barrier
$on@directed method
'no reaction coordinate+
Ho% to sam6le unfa&ora(le
statesG
At high T) (arriers are
easier to o&ercome.
Heat and cool the system
to 6ush it o&er (arriers to
sam6le ne% configurations
Ce6lica E?change
H&ercoming Free Energy Barrier
Qaunch simulations at different
tem6eratures
S%a6 configurations (ased on a
monte carlo criterion
This criterion guarantees that the
lo%est 'target+ tem6erature
]tra^ectory_ sam6les from the
Boltzmann distri(ution.
S%a66ing configurations
effecti&ely im6ro&es sam6ling
H&ercoming Free Energy Barrier
9hen doing 6arallel tem6ering
molecular dynamics) one must ta8e
care in the inter6retation of the
results. A 6arallel tem6ering e?change
is an `un6hysicalW mo&e) and so one
cannot dra% conclusions a(out
dynamics. That is) %hen using 6arallel
tem6ering molecular dynamics) one is
only really doing a form of sam6ling
and not `trueW molecular dynamics.
Deem et al. 3hys . Chem. Chem. 3hys .
- B B V ) R ) S / 4 B S / 4 5
2arcia et al.
3CHTE:$SL Structure)
Function) and Bioinformatics
V/LROSR/B '-BBV+
Ceference
Anna \ohansson
?ray.(mc.uu.se7acalle7mdP6hd7freePenerg.6df
Da&id Mathe%s
htt6L77rna.urmc.rochester.edu7teaching.html
Chris Chi6ot
htt6L77%%%.cirm.uni&@rs.fr7&ideos7-BB57e?6oses7
B-PQeBris7Chi6ot.6df
