# Sat Sep  1 14:15:28 PDT 2007 Kevin Karplus
# WORK IN PROGRESS: NOT READY FOR USE

# The str4 alphabet is a new secondary structure alphabet that can be
#  computed by undertaker.  It is based in part on the successful str2
# alphabet and in part by the n_notor2 alphabet.
#
# There are 3 sections to the alphabet:
#	helix letters, based primarily on hydrogen-bonding patterns
#	strand letters, based on hydrogen bonding and NOtor torsion angles
#	other letters, based mainly on phi-psi angles.

# Helix letters are assigned first.  Any residue that gets a helix
# letter cannot be subsequently relabeled.
# 	If there is an Hbond Oi-2 to Ni+3, with NOtor angle <-133 or >76,
#		then assign letter I	(pi helix).
#	    If there is also a normal helical  Oi-2,Ni+2, then assign letter J.
#`			
#	If there is an Hbond Oi-2 to Ni+1, then assign letter G (3-10 helix or turn).
#	    If there is ALSO Oi-2 to Ni+2, then assign letter F instead.
#		The combined 3-10 and alpha hbonds are most commonly found at
#			the C-terminal end of a helix.
#
#	If there is an Hbond Oi-2 to Ni+2, then assign letter H (helix).
#
#	Note: F,H,J all mean that there is an Oi-2,Ni+2 Hbond.
#	Every separation=4 Hbond results in exactly one F,H, or J being labeled.
#
#	There is some non-reversability in this alphabet, as HJ is
# 	common, but JH is never seen.
#
#	Note that all the helix leter definitions are based on what
# 	happens at Oi-2.  Other definitions were considered, but this
#	seemed to provide the cleanest assignments with common
# 	multiple H-bond possibilities.
#
#   Strand letters are assigned next, based solely on Hbond patterns.
#	Strands are separated by parallel, antiparallel, mixed
#		edge strands are further separated by bonded/unbonded
#		antiparallel bonded residues are further separated by
# 			NOtor angle
#
#	[Sun Oct  7 2007.  
#	     Removed the torsion angle constraints from the parallel
#            strands.  Also added in the multiple hydrogen bonds for
#	     strands.  ]
#
#	For an H-bond to be considered part of a parallel strand,
#	        the sheet-partner of i is j-1, and the
# 		sheet-partner of j is i+1.
#	For an Hbond to be consider part of an antiparallel strand,
#               the sheet-partner of i is j and of j is i.
#
#	If there are multiple Hbonds on the acceptor atom, sidechain
# 		Hbonds are ignored, and the ratio of the probabilities
#		of the hydrogen bonds are checked.  The log of the ratios
#		is taken and if the |logratio| <= 5 and the hydrogen bonds
#               are from  HO_i to HN_j and HN_j+1 they are considered
# 		a multiple strand, otherwise the strongest NO hbond is used 
# 		for classification.
#
#	Let's refer to the Hbonds of residue i as HN_i and HO_i, 
#		if they exist.
#
#	When assigning different letters based on NOtor angle, the
#		threshold is set at <=-17 degrees.  If both HN_i and
# 		HO_i exist, the one with the larger separation is used.
#		Since the bonding partners are usually the same
# 		residue for both in an antiparallel strand, we use
#		HN_i if the bonding partner j>i and HO_i if the bonding 
#		partner j<i.
#
#	Two residues i and j are considered antiparallel if 
#		( (4.5<dist(CAi, CAj) <5.9) AND
#		  (	(HO_i is OiNj and is antiparallel)	#bonded
#		     or (HN_i is OjNi and is antiparallel)	#bonded
#		  )
#		)
#		OR
#		( (3.7<dist(CAi, CAj) <5.4) AND
#		  (   or (HN_i+1 is Oj-1Ni+1 and is antiparallel)	#nonbonded
#		     or (HO_i-1 is Oj+1Ni-1 and is antiparallel)	#nonbonded
#		    )
#		)
#
#	Two residues i and j are considered parallel if  (4.0<dist(CAi, CAj) <5.75)
#		AND (   (HO_i is OiNj+1 and is parallel)	# i-bonded
#		     or (HN_i is Oj-1Ni and is parallel)	# i-bonded
#		     or (HOi-1 is Oi-1Nj and is parallel)	# j-bonded
#		     or (HNi+1 is OjNi+1 and is parallel)	# j-bonded
#		    )
#
#	[ Fri Sep  7 05:45:54 PDT 2007 Kevin Karplus
#		CA-CA distance bounds changed to have different ranges
# 		for the 3 different beta-pair types: parallel,
# 		anti-parallel bonded, and anti-parallel unbonded.
#		The new thresholds exclude about 0.1% of pair
# 		distances at either extreme on the dunbrack-1332 training set.
#	]
#
#	For position i:
#		Find all residues parallel or antiparallel to i.
#		(no assignment if more than 2 such residues).
#
#	    multiple hydrogen bonds on acceptor with |log ratio| < 5 and 
#             HO_i has bond to HN_j and HN_j+1:
#		assign base W.
#
#	    parallel:
#		assign P if two residues, both parallel.
# 	    anti-parallel:
#		assign A or B, if two residues, both antiparallel,
# 			using the NOtor angle for the larger separation of HO_i and HN_i
#	    mixed:
#		assign M if two residues, parallel i bonded, and antiparallel unbonded.
#		assign K or L, if two residues, parallel, j bonded and antiparallel bonded,
# 			using the NOtor angle for the larger separation of HO_i and HN_i
#
#	Edge strands---exactly one residue (j) parallel or antiparallel.
#		Assign	C or D, if j antiparallel bonded,
# 			using the NOtor angle for the larger separation of HO_i and HN_i
#		Assign E, if j antiparallel unbonded,
#		Assign Q, if j parallel i-bonded.
#		Assign R, if j parallel j-bonded.
#		
#
# After all the residues that can be classified with helix and strand
# Hbonds have been classified, the remaining residues (which generally
# have no Hbonds or weird Hbonds) are classified by local backbone geometry.
#

# The other local-structure alphabets we have used (pb, bys, alpha)
# are a bit too big to include in their entirety.  For bys and alhpa,
# we can lump together somewhat similar classes to reduce the total
# number, but reclustering pb seems difficult.  The alpha alphabet
# includes a bit more context than the bys alphabet, though not as
# much as the pb alphabet.

# For bys, we might use 7 classes (2.5 bits):
#	Bys	approx fraction of unassigned
#	H	0.34
#	P	0.22
#	E	0.13
#	G	0.10
#	TL	0.06
#	DN	0.06
#	other	0.08
# This leaves a lot of the residues in a single class.

# For alpha, we might use 5 classes (2.0 bits):
#	alpha	range		approx fraction
#	AE	165 to -136	0.22
#	BCD	-136 to -17	0.23
#	FGHI	-17 to 85	0.40
#	ST	85 to 165	0.13
#	X	undef		0.03

# We could tweak the boundaries a bit, to make the classes sizes more
# uniform, but these correspond fairly well to natural boundaries.

# The estimates of the frequencies of the classes may be way
# off, because they were estimated from the residues that had
# multiple, no, or sidechain H-bonds in o_notor2 and n_notor2, but the
# helix residues will be labeled in a different place on the chain, so
# the unlabeled residues will be a different set.

########################################

# We end up with 5 helix labels, 11 strand labels, and 4 loop labels,
#  for a total of 20 letters in the alphabet.

########################################


ClassName = Alphabet
    Name = str4
    IsNucleic = 0
    NormalChars = ABCDEFGHIJKLMPQRSTUVW

    AllMatch = X

    CharName = F GandH	3_10_and_alpha	
    CharName = G 3_10	3_10_helix_or_turn
    CharName = H helix	alpha_helix
    CharName = I pi	pi_helix
    CharName = J HandI	alpha_and_pi

    # strand types

    CharName = A anti_hi	antiparallel_middle_strand_high_notor
    CharName = B anti_lo	antiparallel_middle_strand_low_notor
    CharName = C anti_edge_hi	antiparallel_edge_strand_high_notor
    CharName = D anti_edge_lo	antiparallel_edge_strand_lo_notor
    CharName = E anti_edge_un	antiparallel_edge_strand_unbonded    
    CharName = K mixed_hi	mixed_middle_strand_high_notor
    CharName = L mixed_lo	mixed_middle_strand_low_notor
    CharName = M mixed_parallel	mixed_middle_strand_bonded_parallel    
    CharName = P parallel	parallel_middle_strand 
    CharName = Q para_edge_bonded	parallel_edge_strand_bonded
    CharName = R para_edge_un	parallel_edge_strand_unbonded

    # other types, based on alpah CAi-1 CAi CAi+1 CAi+2 torsion angle

    CharName = S other_helical	other_alpha_-17_to_85
    CharName = T other		other_alpha_85_to_165
    CharName = U other_str1	other_alpha_165_to_-136
    CharName = V other_str2	other_alpha_-136_to_-17

    # other types, bifurcated Hbond (|log(ratio)| of hbond probabilities 
    # < 5 and seperation between the two residues is one.)

    CharName = W bifurcated	bifurcated_multiple_hbonds

    # If none of the Hbond definitions assign a code, and the residue does
    # not have an alpha torsion angle, assign X.    

EndClassName = Alphabet

# After looking at str4, we developed str5.  str5 is similar to str4, with 
# a few labels merged and a few more created.

# The torsion angle distinction from str4 seems to not be that useful.  
# The anti-parallel distinctions for torsion angle were merged.

# The 3 and 4 hydrogen bonds and the 4 and 5 hydrogen bonds seem to not 
# be that conserved, so the 3 and 4 hydrogen bonds were merged into separation
# 3 hydrogen bonds, and the separation 4 and 5 hydrogen bonds were merged
# into separation 5 hydrogen bonds.

# Some of the results from strand_sep also found their way into the str5
# alphabet.  str5 uses the same coil distinctions from strand_sep, splitting 
# by NO_dist2 and r_rot.  It also has another class for other helical for 
# the end of helices.

# Helical amino acids are split into to classes, long hydrogen bonds (>=2.955)
# and short hydrogen bonds (<2.955).  

########################################

# We end up with 5 helix labels, 8 strand labels, and 5 loop labels,
#  for a total of 20 letters in the alphabet.

########################################


ClassName = Alphabet
    Name = str5
    IsNucleic = 0
    NormalChars = ABCDEFGHIJKLMPQRWYZ

    AllMatch = X

    # Helix types

    CharName = G 3_10	3_10_helix_or_turn
    CharName = H helix_long	alpha_helix_long_hydrogen_bonds
    CharName = I helix_short	alpha_helix_short_hydrogen_bonds
    CharName = J pi	pi_helix
    CharName = K other_helical	other_helical	

    # strand types

    CharName = A anti_middle	antiparallel_middle_strand
    CharName = Y anti_edge_bonded	antiparallel_edge_bonded
    CharName = Z anti_edge_un	antiparallel_edge_strand_unbonded    
    CharName = L mixed_anti	mixed_middle_strand_anti
    CharName = M mixed_parallel	mixed_middle_strand_parallel    
    CharName = P parallel	parallel_middle_strand 
    CharName = Q para_edge_bonded	parallel_edge_strand_bonded
    CharName = R para_edge_un	parallel_edge_strand_unbonded

    # other types, based on NO_dist2 and r_rot

    CharName = B NOdist_low_rot_r_low	NOdist_low_rot_r_low
    CharName = C rot_r_mid		rot_r_mid
    CharName = D NOdist_low_rot_r_high	NOdist_low_rot_r_high
    CharName = E NOdist_high_rot_r_low	NOdist_high_rot_r_low
    CharName = F NOdist_high_rot_r_high	NOdist_high_rot_r_high

   # other types, bifurcated Hbond (|log(ratio)| of hbond probabilities 
    # < 5 and seperation between the two residues is one.)

    CharName = W bifurcated	bifurcated_multiple_hbonds

    # If none of the Hbond definitions assign a code, and the residue does
    # not have an alpha torsion angle, assign X.    

EndClassName = Alphabet