Ma­chine learn­ing tool can pre­dict vi­ral reser­voirs in the an­i­mal king­dom

Tehran Times - - SCIENCE -

Many deadly and newly emerg­ing viruses like Ebola and Zika cir­cu­late in wild an­i­mal and in­sect com­mu­ni­ties long be­fore spread­ing to hu­mans and caus­ing se­vere dis­ease. How­ever, find­ing these nat­u­ral virus hosts – which could help pre­vent the spread to hu­mans – cur­rently poses an enor­mous chal­lenge for sci­en­tists.

Now, a new ma­chine learn­ing al­go­rithm has been de­signed to use vi­ral genome se­quences to pre­dict the likely nat­u­ral host for a broad spec­trum of RNA viruses, the vi­ral group that most of­ten jumps from an­i­mals to hu­mans.

The new re­search, led by the Univer­sity of Glas­gow and pub­lished to­day in Sci­ence, sug­gests this new tool could help in­form pre­ven­tive mea­sures against deadly dis­eases. Sci­en­tists now hope this new ma­chine learn­ing tool will ac­cel­er­ate re­search, sur­veil­lance and dis­ease con­trol ac­tiv­i­ties to tar­get the right species in the wild, with the ul­ti­mate aim of pre­vent­ing deadly and dan­ger­ous viruses reach­ing hu­mans.

Di­verse viruses

Find­ing an­i­mal and in­sect hosts of di­verse viruses from their genome se­quences can take years of in­ten­sive field re­search and lab­o­ra­tory work. The de­lays caused by this mean that it is dif­fi­cult to im­ple­ment pre­ven­tive mea­sures such as vac­ci­nat­ing the an­i­mal sources of dis­ease or pre­vent­ing dan­ger­ous con­tact be­tween species.

Re­searchers stud­ied the genomes of over 500 viruses to train ma­chine learn­ing al­go­rithms to match pat­terns embed­ded in the vi­ral genomes to their an­i­mal ori­gins. These mod­els were able to ac­cu­rately pre­dict which an­i­mal reser­voir host each virus came from, whether the virus re­quired the bite of a blood-feed­ing vec­tor and, if so, whether the vec­tor is a tick, mos­quito, midge, or sand­fly.

Next, re­searchers ap­plied the mod­els to viruses for which the hosts and vec­tors are not yet known, such as Crimean Congo Hem­or­rhagic Fever, Zika and MERS. Model pre­dicted hosts of­ten con­firmed the cur­rent best guesses in each field.

Sur­pris­ingly though, two of the four species of Ebola which were pre­sumed to have a bat reser­voir, ac­tu­ally had equal or stronger sup­port as pri­mate viruses which could point to a non-hu­man pri­mate, rather than bat, source of some Ebola out­breaks.

Piece of in­for­ma­tion

Dr. Daniel Stre­icker, the se­nior au­thor of the study from the MRC-Univer­sity of Glas­gow Cen­ter for Virus Re­search, said: “Genome se­quences are just about the first piece of in­for­ma­tion avail­able when viruses emerge, but un­til now they have mostly been used to iden­tify viruses and study their spread.

“Be­ing able to use those genomes to pre­dict the nat­u­ral ecol­ogy of viruses means we can rapidly nar­row the search for their an­i­mal reser­voirs and vec­tors, which ul­ti­mately means ear­lier in­ter­ven­tions that might pre­vent viruses from emerg­ing all to­gether or stop their early spread.”

Dr. Pete Gard­ner from Well­come’s In­fec­tion & Im­muno­bi­ol­ogy team said: “Healthy an­i­mals can carry viruses which can in­fect peo­ple caus­ing dis­ease out­breaks. Find­ing the an­i­mal species is of­ten in­cred­i­bly chal­leng­ing, mak­ing it dif­fi­cult to im­ple­ment pre­ven­ta­tive mea­sures such as vac­ci­nat­ing an­i­mals or pre­vent­ing an­i­mal con­tact.

“This im­por­tant study high­lights the pre­dic­tive power of com­bin­ing ma­chine learn­ing and ge­netic data to rapidly and ac­cu­rately iden­tify where a dis­ease has come from and how it is be­ing trans­mit­ted. This new ap­proach has the po­ten­tial to rapidly ac­cel­er­ate fu­ture re­sponses to vi­ral out­breaks.”

The re­searchers are now de­vel­op­ing a web ap­pli­ca­tion that will al­low sci­en­tists from any­where in the world to sub­mit their virus se­quences and get rapid pre­dic­tions for reser­voir hosts, vec­tors and trans­mis­sion routes.

Newspapers in English

Newspapers from Iran

© PressReader. All rights reserved.