2021 update

You may prefer the python implementation of most of these kernels available here : https://github.com/jajupmochi/graphkit-learn

ChemoKernel

C++ Library to compute graph kernels on molecular graphs. Mainly my PhD work.

packages required :

• libx11-dev
• cmake
• liblapack-dev

To Compile :

• cd GraphKernels
• mkdir build
• cd build
• cmake ../
• make
• cd ../../
• mkdir build
• cd build
• cmake ../Molecules
• make

References

This library is mainly an implementations of these papers:

• Benoit Gaüzère, Pierre-Anthony Grenier, Luc Brun and Didier Villemin. Treelet kernel incorporating cyclic, stereo and inter pattern information in chemoinformatics. In Pattern Recognition, 2014.
• Benoit Gaüzère, Luc Brun and Didier Villemin. Two new graphs kernels in chemoinformatics. In Pattern Recognition Letters, 2012.
• Benoit Gaüzère, Luc Brun, Didier Villemin. Graph kernel encoding substituents' relative positioning . In International Conference on Pattern Recognition (ICPR), 2014.
• Benoit Gaüzère, Luc Brun, Didier Villemin. Relevant Cycle Hypergraph Representation for Molecules . In 9th IAPR-TC15 Graph Based Representations in Pattern Recognition , 2013.
• Benoit Gaüzère, Luc Brun, Didier Villemin and Myriam Brun. Graph Kernels Based on Relevant Patterns and Cycle Information for Chemoinformatics. In International Conference on Pattern Recognition (ICPR), 2012.
• Benoit Gaüzère, Luc Brun, Didier Villemin and Myriam Brun. Graph Kernels: Crossing Information from Different Patterns Using Graph Edit Distance. In Structural, Syntactic, and Statistical Pattern Recognition (S+SSPR), 2012.

It allows also to compute others graph kernels.

Disclaimer

This is a dirty PhD code. It may bug without any warning. This is published for experiments reproduction purpose.

The Library

This library allows to compute treelet kernel in its various forms. It take datasets as ct (ChemDraw Connection Table) format, as available here

To use a kernel, you may be interested in the following bin files:

• moleculesTreeletData : Extract treelet distribution from a dataset file. Example : $>./bin/moleculesTreeletData ~/work/Datasets/Acyclic/ dataset_bps.ds -o data

This will create 4 files :

• data_treelets_types : Cumulative sum of number of treelets for each kind of pattern

• data_treelets_codes : Canonical codes associated to each treelet

• data_y : The value to predict associated to each graph of a dataset

• data_treelets_distrib : The number of occurences of each treelet for each graph

• ./bin/moleculesComputeGramMatrice ~/work/Datasets/Acyclic/ dataset_bps.ds -gram gram -k 6 Compute a Gram Matrix using treelet kernel and store it in "gram" file. This file is a plain text files encoding a NxN Matrix with kernel values. You can then use it with any SVM or kernel machine.

• ./bin/moleculesComputeGramMatrices ~/work/Datasets/Acyclic/ dataset_bps.ds -k 0 -K 0 -o data Same as previous but a Gram Matrix is provided for each treelet. This is required as an input for Multiple Kernel Learning framework.

• ./bin/moleculesRegression10percent /home/bgauzere/work/Datasets/Acyclic/ dataset_bps.ds -k 6 -s 0.5 -K 0 -a 0.01 Compute a 10-fold cross validation for a regression problem with various parameters. This set of parameters is optimal for Acyclic dataset. Check the output at the end of this file.

• bin/moleculesClassification Same but for classification problems

If any problem, email at benoit.gauzere at gmail . com

Appendix

$>bin/moleculesRegression10percent /home/bgauzere/work/Datasets/Acyclic/ dataset_bps.ds -k 6 -s 0.5 -K 0 -a 0.01

Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:139030 Molecule 0 : -24.3281752461511 (-23.7) --> 0.628175246151084° Molecule 1 : 37.8118531342269 (40) --> 2.18814686577306° Molecule 2 : 159.425402411766 (154) --> 5.42540241176633° Molecule 3 : 87.9121640991155 (88) --> 0.0878359008844711° Molecule 4 : 177.212293826742 (173.7) --> 3.5122938267416° Molecule 5 : 80.7560878008165 (81.2) --> 0.443912199183515° Molecule 6 : 132.011347942363 (132.5) --> 0.488652057637353° Molecule 7 : 128.368044308597 (133.6) --> 5.23195569140347° Molecule 8 : 208.94993853767 (211) --> 2.05006146232986° Molecule 9 : 105.718453042034 (106.5) --> 0.781546957966228° Molecule 10 : 151.377458842552 (146) --> 5.37745884255247° Molecule 11 : 152.653291504656 (159) --> 6.34670849534413° Molecule 12 : 123.034542308033 (122.2) --> 0.834542308032951° Molecule 13 : 149.673273592903 (147) --> 2.67327359290283° Molecule 14 : 231.68588195903 (226) --> 5.68588195903047° Molecule 15 : 148.318062258627 (139) --> 9.31806225862738° Molecule 16 : 157.579312184428 (166.5) --> 8.92068781557211° Molecule 17 : 147.956229274723 (162) --> 14.0437707252772° Molecule 18 : 209.985021965291 (215) --> 5.0149780347088° Mean Prediction Time : 9655 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:407354 Molecule 19 : 17.1188578598173 (14) --> 3.11885785981733° Molecule 20 : 35.2252384547334 (34.6) --> 0.625238454733349° Molecule 21 : 157.022467088595 (156) --> 1.02246708859485° Molecule 22 : 77.0963601287867 (83) --> 5.90363987121334° Molecule 23 : 163.635676718576 (165.5) --> 1.86432328142396° Molecule 24 : 87.7791721172915 (93) --> 5.22082788270848° Molecule 25 : 116.233815067419 (123.5) --> 7.26618493258125° Molecule 26 : 116.233532639876 (120.4) --> 4.16646736012407° Molecule 27 : 127.951629999569 (125) --> 2.95162999956902° Molecule 28 : 104.718515304628 (101) --> 3.71851530462773° Molecule 29 : 139.233449352486 (145) --> 5.76655064751421° Molecule 30 : 156.155226405016 (163.5) --> 7.34477359498425° Molecule 31 : 122.706998234051 (112) --> 10.7069982340511° Molecule 32 : 158.799563106421 (165) --> 6.20043689357914° Molecule 33 : 208.868435951635 (215) --> 6.13156404836514° Molecule 34 : 163.373760179689 (162) --> 1.37376017968941° Molecule 35 : 146.595695380612 (146) --> 0.595695380612455° Molecule 36 : 161.08848842402 (162) --> 0.911511575980484° Molecule 37 : 210.647236174079 (216) --> 5.35276382592144° Mean Prediction Time : 9160 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:651766 Molecule 38 : 49.6999106195394 (37.3) --> 12.3999106195394° Molecule 39 : 30.4092781768384 (32) --> 1.59072182316156° Molecule 40 : 156.567779015236 (166) --> 9.43222098476417° Molecule 41 : 103.566914175643 (104.5) --> 0.933085824356539° Molecule 42 : 174.76648719166 (181) --> 6.23351280833981° Molecule 43 : 77.1586380731574 (69) --> 8.15863807315738° Molecule 44 : 115.025633943629 (120.3) --> 5.27436605637082° Molecule 45 : 119.549886212784 (120) --> 0.450113787215912° Molecule 46 : 116.812598609236 (118) --> 1.18740139076445° Molecule 47 : 95.283455569108 (99.3) --> 4.01654443089195° Molecule 48 : 130.425227178775 (141) --> 10.5747728212247° Molecule 49 : 206.931260945931 (229.5) --> 22.5687390540693° Molecule 50 : 96.6454730566092 (106) --> 9.35452694339079° Molecule 51 : 189.675960264429 (188.9) --> 0.775960264428733° Molecule 52 : 181.111475241989 (201) --> 19.8885247580111° Molecule 53 : 174.780249491224 (173) --> 1.78024949122408° Molecule 54 : 177.916802818836 (165) --> 12.9168028188356° Molecule 55 : 174.08849821132 (173.2) --> 0.888498211320382° Molecule 56 : 232.292198423554 (240) --> 7.70780157644555° Mean Prediction Time : 8797 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:882401 Molecule 57 : 107.705220947439 (109.7) --> 1.99477905256141° Molecule 58 : 56.4297349251341 (63) --> 6.57026507486592° Molecule 59 : 169.387392061573 (183) --> 13.6126079384275° Molecule 60 : 101.948311973125 (102) --> 0.0516880268754392° Molecule 61 : 102.301008779956 (99.5) --> 2.80100877995638° Molecule 62 : 86.2004502512818 (86.3) --> 0.0995497487182035° Molecule 63 : 152.396904593896 (145) --> 7.39690459389556° Molecule 64 : 144.411111298084 (137) --> 7.41111129808434° Molecule 65 : 118.637607088911 (117.1) --> 1.53760708891086° Molecule 66 : 111.237976678891 (90) --> 21.2379766788913° Molecule 67 : 171.830681246313 (171) --> 0.830681246312793° Molecule 68 : 142.208269295778 (142) --> 0.208269295778166° Molecule 69 : 111.565147361701 (114.5) --> 2.93485263829855° Molecule 70 : 166.766973708094 (170) --> 3.23302629190636° Molecule 71 : 201.368111579662 (205) --> 3.63188842033816° Molecule 72 : 159.112797831761 (159.5) --> 0.387202168239156° Molecule 73 : 155.223217744827 (159) --> 3.77678225517292° Molecule 74 : 188.471882801062 (186.8) --> 1.67188280106205° Mean Prediction Time : 8659 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:1107560 Molecule 75 : 5.03101639149838 (10.8) --> 5.76898360850162° Molecule 76 : 57.4005341117461 (53.5) --> 3.90053411174607° Molecule 77 : 71.9074928978082 (70.3) --> 1.60749289780817° Molecule 78 : 100.490338650428 (92) --> 8.49033865042831° Molecule 79 : 94.0742217883446 (92.3) --> 1.77422178834459° Molecule 80 : 83.5748561648457 (82) --> 1.57485616484567° Molecule 81 : 148.279948810023 (144.2) --> 4.07994881002324° Molecule 82 : 190.953653287028 (195.8) --> 4.84634671297209° Molecule 83 : 107.703367889279 (107) --> 0.703367889279306° Molecule 84 : 138.707530012616 (137) --> 1.70753001261554° Molecule 85 : 170.599146108412 (166) --> 4.59914610841167° Molecule 86 : 133.791161890264 (125) --> 8.79116189026365° Molecule 87 : 153.652778718362 (151) --> 2.65277871836196° Molecule 88 : 182.873182580313 (178) --> 4.8731825803134° Molecule 89 : 146.445203518336 (151.5) --> 5.0547964816642° Molecule 90 : 156.272636167033 (159.5) --> 3.22736383296737° Molecule 91 : 211.988725399332 (195) --> 16.9887253993316° Molecule 92 : 175.688411914943 (173) --> 2.68841191494269° Mean Prediction Time : 8563 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:1332302 Molecule 93 : 38.941662029324 (39) --> 0.0583379706759715° Molecule 94 : 67.5453677824893 (64.4) --> 3.14536778248925° Molecule 95 : 62.2976697386275 (63.6) --> 1.30233026137248° Molecule 96 : 109.626781930352 (107.4) --> 2.22678193035182° Molecule 97 : 84.2295543238165 (90.1) --> 5.87044567618346° Molecule 98 : 102.588580309409 (103) --> 0.411419690590904° Molecule 99 : 140.319044522254 (142.8) --> 2.48095547774614° Molecule 100 : 177.718209097898 (177.2) --> 0.518209097897568° Molecule 101 : 102.648225789402 (102.5) --> 0.14822578940219° Molecule 102 : 115.387764945301 (114) --> 1.38776494530066° Molecule 103 : 151.115089695928 (155) --> 3.88491030407249° Molecule 104 : 131.040930825716 (132) --> 0.959069174283513° Molecule 105 : 119.002430575814 (128.5) --> 9.49756942418563° Molecule 106 : 159.564452733536 (148.5) --> 11.0644527335355° Molecule 107 : 159.29990716573 (165.5) --> 6.20009283427046° Molecule 108 : 163.833297257609 (155.5) --> 8.33329725760916° Molecule 109 : 208.776634695136 (218) --> 9.22336530486393° Molecule 110 : 179.229192732816 (187) --> 7.77080726718401° Mean Prediction Time : 8480 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:1554796 Molecule 111 : 51.8968378417414 (42) --> 9.89683784174142° Molecule 112 : 86.9802976060384 (84.7) --> 2.28029760603836° Molecule 113 : 52.1855698470543 (52.5) --> 0.314430152945746° Molecule 114 : 123.234348536773 (123.2) --> 0.0343485367725265° Molecule 115 : 78.165711877495 (80.2) --> 2.03428812250499° Molecule 116 : 98.1102462588063 (103.5) --> 5.38975374119369° Molecule 117 : 136.368503834857 (132) --> 4.36850383485717° Molecule 118 : 185.266553760578 (181) --> 4.26655376057801° Molecule 119 : 105.531804810588 (112) --> 6.46819518941189° Molecule 120 : 127.486118916511 (126) --> 1.48611891651119° Molecule 121 : 150.670137798131 (145) --> 5.67013779813055° Molecule 122 : 129.855693552027 (130.5) --> 0.644306447973435° Molecule 123 : 110.277848021086 (109.5) --> 0.777848021085873° Molecule 124 : 168.104780265245 (165) --> 3.10478026524507° Molecule 125 : 150.52337317439 (157) --> 6.47662682561028° Molecule 126 : 141.986773101938 (141) --> 0.986773101937899° Molecule 127 : 246.777639196288 (250) --> 3.22236080371198° Molecule 128 : 173.25259953402 (174) --> 0.747400465980348° Mean Prediction Time : 8440 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:1777887 Molecule 129 : 75.5427115805266 (66.6) --> 8.94271158052659° Molecule 130 : 101.099149026897 (95.5) --> 5.5991490268967° Molecule 131 : 54.662927581659 (59) --> 4.33707241834104° Molecule 132 : 113.828417649685 (112.5) --> 1.32841764968538° Molecule 133 : 77.8693062280687 (82) --> 4.13069377193128° Molecule 134 : 96.6017751925423 (96) --> 0.601775192542348° Molecule 135 : 133.436667638644 (134.2) --> 0.763332361355538° Molecule 136 : 186.072789643 (185.9) --> 0.172789643000158° Molecule 137 : 93.0012038861267 (97.4) --> 4.39879611387329° Molecule 138 : 131.522001456978 (124) --> 7.52200145697796° Molecule 139 : 155.685925398326 (159) --> 3.31407460167406° Molecule 140 : 117.54380834091 (125) --> 7.45619165908954° Molecule 141 : 124.115667000344 (126) --> 1.88433299965583° Molecule 142 : 180.216142969628 (177) --> 3.21614296962804° Molecule 143 : 130.363797521203 (139) --> 8.63620247879729° Molecule 144 : 126.729485223868 (126) --> 0.729485223868267° Molecule 145 : 212.673843502615 (235) --> 22.3261564973848° Molecule 146 : 191.036584101807 (188.5) --> 2.53658410180677° Mean Prediction Time : 8420 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:2002690 Molecule 147 : 131.771664565311 (135) --> 3.22833543468897° Molecule 148 : 89.9335633782254 (92) --> 2.06643662177457° Molecule 149 : 59.7913152334319 (59.5) --> 0.291315233431902° Molecule 150 : 115.184632601919 (118.5) --> 3.31536739808077° Molecule 151 : 90.9088105735942 (91.2) --> 0.291189426405836° Molecule 152 : 113.23978842908 (112) --> 1.23978842907971° Molecule 153 : 143.066453526927 (137) --> 6.0664535269274° Molecule 154 : 172.358353860901 (175.7) --> 3.34164613909877° Molecule 155 : 101.639396807234 (91.5) --> 10.1393968072339° Molecule 156 : 137.538112876012 (140.5) --> 2.96188712398751° Molecule 157 : 141.854642403291 (138) --> 3.85464240329142° Molecule 158 : 116.580220532949 (122) --> 5.41977946705137° Molecule 159 : 146.01558045644 (147) --> 0.984419543560477° Molecule 160 : 165.692901408809 (167) --> 1.30709859119108° Molecule 161 : 158.563981747604 (163) --> 4.43601825239631° Molecule 162 : 147.962404540204 (164) --> 16.0375954597962° Molecule 163 : 191.0966324151 (186.5) --> 4.5966324151002° Molecule 164 : 182.324933202676 (199) --> 16.6750667973235° Mean Prediction Time : 8395 Spectrum Kernel : Intersection KernelGram Matrix computed Learning Time:2226576 Molecule 165 : 144.111585362524 (148.5) --> 4.38841463747562° Molecule 166 : 86.3777698939996 (84.4) --> 1.97776989399962° Molecule 167 : 62.8321864701092 (55.2) --> 7.63218647010916° Molecule 168 : 102.731389204222 (101.5) --> 1.23138920422204° Molecule 169 : 91.7242795866036 (91.5) --> 0.224279586603615° Molecule 170 : 106.888999823352 (104) --> 2.88899982335201° Molecule 171 : 141.342264167235 (139) --> 2.34226416723547° Molecule 172 : 198.541406185377 (186) --> 12.5414061853771° Molecule 173 : 94.5260562022613 (87.6) --> 6.92605620226129° Molecule 174 : 148.262019269227 (157.5) --> 9.23798073077347° Molecule 175 : 145.147244873906 (142) --> 3.14724487390643° Molecule 176 : 133.546473813741 (121) --> 12.5464738137407° Molecule 177 : 155.649958864346 (158) --> 2.3500411356537° Molecule 178 : 190.121291467466 (195) --> 4.87870853253386° Molecule 179 : 162.469799984499 (153.5) --> 8.96979998449919° Molecule 180 : 161.293419941354 (163) --> 1.70658005864649° Molecule 181 : 145.313801441483 (156.5) --> 11.1861985585172° Molecule 182 : 228.02931998909 (228) --> 0.0293199890904248° Mean Prediction Time : 8472

Average error : 4.69482116282133
Standard Deviation : 6.45314284880126
R : 0.990933085827914