def sankoff(tree, possible_labels = ["LL", "RE", "RW", "M1", "M2", "Liv"]):
root = [n for n in tree if tree.in_degree(n) == 0][0]
tree = cmp.set_depth(tree, root)
max_depth = cmp.get_max_depth(tree, root)
tree = cmp.extend_dummy_branches(tree, max_depth)
C = np.full((len(tree.nodes), len(possible_labels)), np.inf)
# create maps betwene names and row/col of dynamic programming array, C
bfs_postorder = [root]
for e0, e1 in nx.bfs_edges(tree, root):
bfs_postorder.append(e1)
node_to_i = dict(zip(bfs_postorder, range(len(tree.nodes))))
label_to_j = dict(zip(possible_labels, range(len(possible_labels))))
# instantiate the dynamic programming matrix at leaves
_leaves = [n for n in tree.nodes if tree.out_degree(n) == 0]
for l in _leaves:
label = tree.nodes[l]["label"]
i, j = node_to_i[l], label_to_j[label]
C[i, j] = 0
C = sankoff_fill_C(tree, root, C, node_to_i, label_to_j)
return C
def compute_transitions_majority_vote(t, meta):
possible_labels = meta.unique()
M = len(possible_labels)
C = np.zeros((M, M))
label_to_j = dict(zip(possible_labels, range(len(possible_labels))))
root = [n for n in t if t.in_degree(n) == 0][0]
t = small_parsimony.assign_labels(t, meta)
t = cmp.set_depth(t, root)
t = assign_majority_vote(t, root)
# now count transitions
for v in nx.dfs_postorder_nodes(t, source=root):
v_lab = t.nodes[v]['label']
i = label_to_j[v_lab]
children = list(t.successors(v))
for c in children:
c_lab = t.nodes[c]['label']
j = label_to_j[c_lab]
C[i, j] += 1
count_mat = pd.DataFrame(C)
count_mat.columns = possible_labels
count_mat.index = possible_labels
return count_mat
def fitch_count(t, meta):
root = [n for n in t if t.in_degree(n) == 0][0]
t = small_parsimony.assign_labels(t, meta)
possible_labels = meta.unique()
t = cmp.set_depth(t, root)
t = small_parsimony.fitch_hartigan_bottom_up(t, root, possible_labels)
bfs_postorder = [root]
for e0, e1 in nx.bfs_edges(t, root):
bfs_postorder.append(e1)
node_to_i = dict(zip(bfs_postorder, range(len(t.nodes))))
label_to_j = dict(zip(possible_labels, range(len(possible_labels))))
L = small_parsimony._N(t, possible_labels, node_to_i, label_to_j)
C = small_parsimony._C(t, L, possible_labels, node_to_i, label_to_j)
M = pd.DataFrame(np.zeros((L.shape[1], L.shape[1])))
M.columns = possible_labels
M.index = possible_labels
# count_mat: transitions are rows -> columns
for s1 in possible_labels:
for s2 in possible_labels:
M.loc[s1, s2] = np.sum(C[node_to_i[root], :, label_to_j[s1], label_to_j[s2]])
return M
def naive_fitch(t, meta):
root = [n for n in t if t.in_degree(n) == 0][0]
t = small_parsimony.assign_labels(t, meta)
possible_labels = meta.unique()
t = cmp.set_depth(t, root)
label_to_j = dict(zip(possible_labels, range(len(possible_labels))))
M = small_parsimony.draw_one_solution(t, possible_labels, label_to_j)
M = pd.DataFrame(M)
M.columns = possible_labels
M.index = possible_labels
return M
def fitch_hartigan(tree):
"""
Runs the Hartigan algorithm (a generalization to Fitch's algorithm on nonbinary trees)
on tree given the labels for each leaf. Returns the tree with labels on internal node.
"""
_leaves = [n for n in tree if tree.out_degree(n) == 0]
root = [n for n in tree if tree.in_degree(n) == 0][0]
# form candidate set of labels for each internal node
S = np.unique(np.concatenate([tree.nodes[l]['S1'] for l in _leaves]))
tree = cmp.set_depth(tree, root)
tree = fitch_hartigan_bottom_up(tree, root, S)
tree = fitch_hartigan_top_down(tree, root)
return tree
def compute_transitions_naive(t, meta):
root = [n for n in t if t.in_degree(n) == 0][0]
t = assign_labels(t, meta)
possible_labels = meta.unique()
t = cmp.set_depth(t, root)
t = fitch_parsimony.fitch_bottom_up(t, root)
label_to_j = dict(zip(possible_labels, range(len(possible_labels))))
t = fitch_parsimony.reconcile_fitch(t)
count_mat = fitch_parsimony.draw_one_fitch_solution(t, possible_labels, label_to_j)
count_mat = pd.DataFrame(count_mat)
count_mat.columns = possible_labels
count_mat.index = possible_labels
return count_mat
def compute_transitions(t, meta, count_unique = False): root = [n for n in t if t.in_degree(n) == 0][0] t = assign_labels(t, meta) possible_labels = meta.unique() t = cmp.set_depth(t, root) t = fitch_parsimony.fitch_bottom_up(t, root) bfs_postorder = [root] for e0, e1 in nx.bfs_edges(t, root): bfs_postorder.append(e1) node_to_i = dict(zip(bfs_postorder, range(len(t.nodes)))) label_to_j = dict(zip(possible_labels, range(len(possible_labels)))) root_labs = t.nodes[root]['label'] #if 'LL' not in root_labs: # t.nodes[root]['label'] = np.concatenate((root_labs, ['LL'])) t = fitch_parsimony.reconcile_fitch(t) L = fitch_parsimony.count_opt_solutions(t, possible_labels, node_to_i, label_to_j) obs_transitions = defaultdict(list) C = fitch_parsimony.count_num_transitions(t, L, possible_labels, node_to_i, label_to_j, count_unique = count_unique) count_mat = pd.DataFrame(np.zeros((L.shape[1], L.shape[1]))) count_mat.columns = possible_labels count_mat.index = possible_labels # count_mat: transitions are rows -> columns for s1 in possible_labels: for s2 in possible_labels: count_mat.loc[s1, s2] = np.sum(C[node_to_i[root], :, label_to_j[s1], label_to_j[s2]]) return count_mat